Page 5 – The DIY Life

Pironman 5 NVMe Raspberry Pi 5 Case Review

Michael Klements

-

July 31, 2024

Pironman 5 NVMe Raspberry Pi 5 Case Review

Today we’re going to be taking a look at new the Pironman 5 case by Sunfounder. This case has been designed to house a Raspberry Pi 5 along with an NVMe storage drive.

It’ll set you back almost $80, but it has quite a lot to offer. So let’s see how easy it is to put together, how the Pi and NVMe drive perform in it and whether it’s worth the money.

This is the second generation of Pironman case, although it doesn’t look all that similar to the first. The first generation was made for the Pi 4, which didn’t have a PCIe port. This case takes advantage of the port to add an NVMe drive to the Pi without the need for an external USB jumper. They’ve also shifted from a two-opposite clear side panel design to a wrap-around clear panel design.

Here’s my unboxing and review video, read on for the write-up;

Where To Get The Pironman 5 Case

Pironman 5 Case – Buy Here
Raspberry Pi 5 – Buy Here
NVMe Storage Drive – Buy Here
Pi 5 USB C Power Supply – Buy Here

Unboxing and Assembly

The case comes in a fairly large white branded box. Inside it is the aluminium case shell packed full of the case components, including fans, cooler, expansion boards and mounting hardware.

Oddly, I did get a random open and empty sleeve in mine. I don’t know if that means something is missing or if it just found its way into the box but I guess I’ll find out when assembling it.

You can see from the spread that there are a lot of components to this case. Like with the previous Pironman case, it’s going to take more than a couple of minutes to put it together.

Assembling The Pironman 5 Case

To assemble the case, you start by adding standoffs to one enclosure half.

You then plug some of the carrier boards into the Pi, then mount the assembly into the case.

The power button and cooler go in next. They supply pads for the WiFi chip and power circuitry which is a bit different. Previously this cooler covered the RAM chips, USB and Ethernet controller.

Next, we install the NVMe drive.

The NVMe drive carrier board supports multiple drive sizes from 2230 to 2280. I’m going to be using a 2280-size Lexar drive in the case. This is not all that fast as far as NVMe drives go, but it runs at gen-3 speeds, so it should get quite close to the maximum speed that the Pi can handle.

We then attach the fans to the back panel.

With that done, we can mount the GPIO expansion board with the OLED display.

Then screw the two aluminium case halves together and stick the display onto the front panel.

To finish it off, we add the clear acrylic panels.

Assembly is relatively easy. The case comes with a good step-by-step illustrated instruction sheet and the hardware is labelled well too. It took me about 30 minutes to have the case complete and ready to boot up.

Pironman 5 Software and Web Dashboard

Booting up the Pi is done by pressing the power button on the front of the case.

To get the display and lighting to work, we need to install some additional software. This is a fairly simple GitHub install by entering the following commands in the terminal:

cd ~
git clone https://github.com/sunfounder/pironman5.git
cd ~/pironman5
sudo python3 install.py

The script took about a minute to install and I didn’t run into any issues with it.

The fans turn off and the OLED display comes on when the installation completes, but it still recommends a reboot. The OLED display on the front shows the CPU usage and temperature, the computer’s IP address as well as the RAM and storage capacity.

Their software also has a really nice web dashboard, accessible by entering the IP address and port 34001 into the browser.

http://<ip>:34001

This dashboard allows you to see your system stats, plot graphs of the stats, see logs and access some of the case script settings.

You can also change the style and colour of the RGB lighting, adjust its pulse mode or set it to cycle through different colours.

Testing The Case’s Cooling Performance

One of Sunfounder’s claims about the case is superior cooling. I personally use one of these tower-style coolers on my Pi 4s and 5s running in my 3D-printed cases. So I’ll use that as a baseline and see if the additional fans and this slightly different cooler base make any difference to the thermals.

I’m also interested to see how loud the case is when it is running as we’ve got three 40mm fans in it. The back two are not PWM fans, so they’re either on at full speed or they’re off. The default setting is for them to turn on when the CPU reaches 60°C, but this can be adjusted.

I’m going to adjust them to have the back two fans running at full speed for the stress test to see how it compares to my case’s thermals.

My case has a single 40mm PWM fan and I’ll set this to 100% for the comparison as well.

I’m going to be using CPU Burn to apply full load to the CPU and we’ll leave that running to see what temperature it stabilises at. I doubt we’ll get anywhere near thermal throttling with this cooling setup.

To install CPU Burn, we enter the following command into the terminal;

wget https://raw.githubusercontent.com/ssvb/cpuburn-arm/master/cpuburn-a53.S
gcc -o cpuburn-a53 cpuburn-a53.S

Then to run the test, we enter the following command;

while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 5; done& ./cpuburn-a53

At the stock CPU frequency of 2.4GHz, at idle, the CPU temperature started at 25°C. I left the test running for a little under 30 minutes and it stabilised at 52°C. So we’ve got a delta of about 27°C, which is really good. This means you’ve got a lot of headroom for overclocking.

The fans are not as noisy as expected, but they would be annoying if you had this on a desk next to you and they were running the whole time. I’ve put an audio clip of them running into my video at the beginning of the post. Thankfully, having them only turn on at 60°C means that they’re off most of the time and only come on when you put a demanding load on the Pi.

With the fans still on, the Pi’s CPU drops back down by 10°C in about 5 minutes.

I ran the same test on my case. This time the starting temperature was 22°C, so 3°C below the Pironman case. The temperature reached equilibrium much faster but I left it running for 30 minutes too. It stabilised at 36°C. So we had a delta of about 14°C, which is significantly better than on the Pironman 5 case.

I think this is mainly down to the airflow path. My case design has airflow straight across the cooler and out some large ventilation holes on the opposite side.

The Pironman case doesn’t really have air inlet vents and the fans at the back are each pushing air through a restrictive dust filter. So although it’s got two more fans than my setup, they aren’t working as effectively.

Testing The NVMe Drive Speed

To test the NVMe drive speed, I’m using James Chambers Pi Benchmarks script. This script favours random read-write performance, so is a good representation of how an OS would be using the drive.

To install and run the test, we enter the below command into the terminal;

sudo curl https://raw.githubusercontent.com/TheRemote/PiBenchmarks/master/Storage.sh | sudo bash

Over 3 consecutive tests, I got an average score of 51,963.

This is a similar score to what I got using the Piromoni NVMe base with this drive on my setup, I got 51,902 and an average of 51,865. So that’s a good indication that there are no issues with the drive adaptor.

It is quite a bit slower than the speeds I got in my recent NVMe hat comparison, but that was done with a Sabrent Rocket drive which is much faster than the Lexar drive I’m currently using.

Final Thoughts On The Pironman 5 Case

Overall I think the case looks great and provides some nice functionality over the stock Pi. I particularly like the full-size HDMI ports over the stock micro HDMI ports and the fact that the cables run out the back of the case rather than on the back and side.

There are also a few good improvements on the original design, assembly is quite a bit easier and it doesn’t rely on the side panels to hold the metal case components in place. So you can have the side panel removed to work on the Pi without compromising on the case’s structural integrity.

In terms of size, it is slightly larger than my 3D-printed case. It measures 112 x 117 mm and is 79mm wide, but they’ve managed to cram a lot of features into the small space.

It’s obvious that Sunfounder have tried to make this the best case for a Raspberry Pi 5. It comes with quite a high price tag, but in terms of value for money, it is fairly good. You can pick this up and you don’t need to worry about getting a separate cooler, NVMe hat, fans or an OLED display, and you still have access to the core Pi’s functionality like its GPIO pins.

It also comes with some nice finishing touches, like labels for the ports, and includes plenty of additional screws and mounting hardware so you’re covered if you lose some of them.

I never found anything missing from the case hardware so I assume that the empty sleeve I found made its way into the package by accident.

If I had to pick out some things to be critical about, I probably would have made the back fans PWM controllable as well. They would then run a bit quieter, although being able to turn them off most of the time in software partially helps with this already.

The dust filters on the back fans are also unnecessary as these are exhaust fans, so you’re filtering dust out that would be leaving the case. These would be better positioned onto some inlet vents and doing so would improve airflow through the exhaust vents.

Let me know what you think of the Pironman 5 case in the comments section below.

The New Beelink GTi 14 Ultra Has A PCIe x8 Slot

Michael Klements

-

July 17, 2024

3

The New Beelink GTi 14 Ultra Has A PCIe x8 Slot

Beelink got my attention two weeks ago when they reached out about a new mini PC that they were about to launch, the GTi 14 Ultra, which has a built-in full-size PCIe gen 4 x8 slot.

One of the biggest weaknesses in these mini PCs is graphics performance. They typically rely on an integrated GPU, and while some have reasonably powerful integrated graphics, they don’t come close to having a dedicated GPU.

I recently showed a way to get around this by taping into an M.2 port on a PC with an Oculink adaptor. This worked fairly well but was limited by the single PCIe lane and looked like a bit of a hack job. Even in a custom 3D-printed case. So I’m super excited to try out this new PC with a fully accessible PCI slot built in.

Here’s my video review of the GTi 14 Ultra, read on for my written review;

Where To Buy The Beelink GTi 14 Ultra

Beelink GTi 14 Ultra – Buy Here

Tool & Equipment Used

USB-C Pencil Screwdriver – Buy Here
Infiray P2 Pro Thermal Camera – Buy Here

Unboxing The GTi 14 Ultra

The GTi 14 Ultra comes in a white Beelink branded box with minimal text on it. Let’s get it unboxed and see what is included.

There are two versions of this PC that are going to be available. This is the less powerful Intel Core Ultra 7 version and it also comes in an Intel Core Ultra 9 version.

Removing the lid, we’ve got the GTi 14. It’s aluminium housing is protected by a matt plastic film.

Underneath the computer, we’ve got two cables. It looks like it’s got the power supply integrated into the PC enclosure, since we’re only provided with a power lead. This is a bit different to typical Mini PC designs which usually relay on a 19V laptop-style power supply to power them.

So included in the box is the mini PC, a power cable, an HDMI cable and a short user manual.

The GTi 14 Ultra is a fair bit larger than most mini PCs I’ve tried previously, but that’s due to the integrated power supply, the full size PCIe port and the need for better cooling on the more powerful CPU.

You definitely get Mac Mini vibes from it.

Another thing worth mentioning is that this mini PC doesn’t come with a VESA mount to mount it onto the back of a monitor. It’s designed to be placed onto a desk.

GTi 14 Ultra’s Specifications and Interfaces

The GTi 14 has an Intel Ultra 7 155H Processor, which is essentially a CPU, GPU and NPU all on a single chip. This is a mobile processor with 24MB of cache and 16 processor cores. It’s got 6 performance cores that can run at up to 4.8Ghz and 8 efficiency cores that can run at up to 3.8Ghz.

It has an integrated Intel Arc GPU with a maximum frequency of 2.25GHz and this supports hardware-based ray tracing, so we should be able to run some games on it to try that out.

It’s got 32GB of DDR5 RAM running at 5600MHz and a 1TB NVMe SSD.

In terms of IO, on the front, we’ve got a USB 3.2 port, a full-size SD card slot, a USB C port, a 3.5mm audio jack, a power indicator LED and a power button. The power button also has an integrated fingerprint sensor on it. The four holes along the top are for a microphone array suited to voice recognition.

The two sides have nothing on them, they’re just bare aluminium.

On the back, we’ve got our AC input, another USB C port (this one is a Thunderbolt 4 port), another 3.5mm audio jack, an HDMI port that can do 4K 60Hz, a DisplayPort that can do 4K 144Hz, two more USB 3.2 ports above two 2.5G Ethernet ports, and another two USB 3.2 ports alongside them.

In addition to the 2.5G Ethernet ports, it’s also got WiFi 7 and Bluetooth 5.4.

The last port, and the one I’m most excited to try out, is the PCIe gen 4 x8 slot, which is accessible through the bottom of the PC under a thin rubber cover.

Cooling is achieved by drawing air in through the mesh on the bottom and then exhausting it through these vents at the back.

Taking A Look Inside The GTi 14 Ultra

Next, let’s open it up and take a look inside. Four screws hold the bottom cover in place. With those removed, we can see two integrated speakers and a dust filter to protect the components.

Bottom Cover Removed, Speakers & Dust Filter

Under the filter is the speaker assembly and our power supply. These make it a bit more difficult to get to the RAM and SSD, but I’m going to go ahead and remove them so we can get a good look at them.

If you try this yourself, you need to remove the speaker assembly first. Once that has been removed, then remove the power supply. This is quite an interesting design. It gets AC power from the port at the back and then feeds 19V DC into the motherboard through these two standoffs. Lastly, remove the cover plate underneath them and then you’re in.

Dust Filter Removed, Speakers and Power Supply

Then we can see our RAM and storage. The RAM is in a dual-channel configuration and is upgradable to a maximum of 96GB. There is also a second slot to add another M.2 NVMe storage drive. Alongside that is a removable M.2 WiFi adaptor and then our PCIe port. It looks like we’ve got a x8 and a x1 slot alongside each other.

First Boot and Benchmarking

Next, let’s close up the bottom cover and try boot it up. The GTi 14 Ultra comes with a clean install of Windows 11 Pro on it. Once set up, you can log in using the fingerprint sensor on the front, which is impressively fast.

I’ve had a look through the software and there doesn’t seem to be any preinstalled bloatware or spyware. You need to be careful buying mini PCs from Amazon or Aliexpress as they’re often filled with nefarious software. Beelink are a reputable brand and have been around for a while so they’re a safe bet.

Opening up the performance monitor we can see our CPU is an Intel Core Ultra 7 and it’s running at a base speed of 3GHz. We’ve got our 32GB of RAM running at 5600Mhz and our 1TB storage drive shows up as well. Our GPU is an Intel Arc and it’s sharing 16Gb of RAM.

Next, we’ll run two benchmarks on it. The first is Geekbench to test CPU and GPU performance.

The CPU benchmark took 4:30 minutes to complete and the fan was surprisingly quiet throughout the benchmark. You could hear it running, but it’s nowhere near as loud as some other mini PCs I’ve tested. We get a single core score of 2,270 and a multicore score of 11,834. So single core scores are fairly average, but the multicore score is good.

The GPU benchmark took just under 2 minutes and we got a score of 37,460. This is very good for an integrated GPU.

Next, let’s run Furmark to test the computer’s GPU and thermals. Like with the Geekbench benchmark, the fan came on almost immediately but wasn’t all that loud for the duration of the test.

On completion, we get a score of 1,920. Unfortunately, the GPU temperature was unable to be recorded.

Gaming On The Stock GTi 14 Ultra

Now that we’ve run some benchmarks, let’s try running some games on it. We’ll start with Counterstrike 2.

It was handling the Home Screen pretty well so I set the graphics settings to Very High. In game, we get around 50-60fps, which is very playable. This is very good for a PC without a dedicated GPU and all settings on Very High. We’ll see how it compares when we add a GPU through the PCIe slot later.

Next, I tried running Doom Eternal. It had a bit of a freak-out when starting the game but it ran well after fixing the aspect ratio and setting the resolution back to 1080P.

I had all graphics settings on Ultra Nightmare and Ray Tracing turned off. In game, I was getting a little over 60fps fairly consistently, which is also really good.

Turning ray tracing on didn’t make a significant difference to performance, we lost about 10fps.

So for 1080P gaming on this mini PC, you really don’t need an external GPU. It does very well with the integrated Intel Arc graphics. Both games were easily playable, but we’re going to try to see how the PCIe port performs in any case.

So let’s plug our GPU into the GTi14 and get the AMD drivers installed.

Gaming On The GTi 14 Ultra With An External GPU

I 3D printed a new bottom cover for the GTi 14 Ultra which has some mounting points for a PCIe riser to plug the GPU into. I’m using a XFX Radeon RX 6600.

Adding The GPU To The GTi 14 Ultra

From the photographs on Beelink’s product page, it looks like they plan on selling an external GPU dock which will plug into the GTi 14 Ultra’s PCIe port. This isn’t yet available so I had to make another plan.

I 3D printed a new bottom cover for the PC. This picks up on the same cover plate screw holes but is offset with some M2.5x12mm standoffs.

GTi 14 Ultra Back Cover Download

I then plugged a Coolermaster PCIe riser into the PCIe port and mounted the female port onto the 3D-printed cover plate with some M3x8mm screws into M2.5 brass inserts.

To power the GPU, I used a 12V 10A power supply which I soldered to an 8-pin plug that I salvaged off a Molex to 8-pin adaptor.

Gameplay With The GTi 14 Ultra

In Counterstrike, with graphics again set to Very High, we’re now getting over 200 fps. This is 3 times what we were getting on the integrated GPU. The game is responsive and the PC seems to be running a bit cooler, the fan is noticeably quieter.

Next, let’s try Doom Eternal. Again with the same Ultra Nightmare graphics settings and Ray Tracing turned off, we’re getting over 150fps. This is about 2.5 times better than on the integrated GPU. I also noticed significantly faster load times with this setup.

Testing Power Consumption & Speakers

With the PC returned to its stock state with nothing plugged into the PCIe port, the GTi 14 Ultra uses around 30W when idle on the desktop. It maxes out at a little over 80W with the GPU and CPU being utilised during gaming.

Power Consumption During Gameplay — Power Consumption Idle

The built-in speakers are a nice inclusion. They lack base because of their size but don’t sound terrible. They’re about on par with a mid-range laptop. You can hear a sample of the audio in my Youtube video.

Final Thoughts On The Beelink GTi 14 Ultra

Overall I think that this is a really awesome mini PC. It’s ultra-portable and having the ability to plug a GPU directly into it gives you the flexibility to use it for some fairly demanding gaming when you’ve got a bit more desk space and don’t need to carry it around. It’s also upgradable with non-soldered RAM and an additional M.2 port.

I have two criticisms though.

One is that the integrated power supply doesn’t seem to go into a proper dormant or sleep state, even when the PC is completely shut down. In the below images, it had been off overnight and the enclosure was still noticeably warm. The power meter registers about 2.5W with the computer shutdown, so it’s using power for no reason.

The second is the implementation of the PCIe slot. It is very deeply recessed in the enclosure and the access slot through the enclosure is too thin for most standard risers. I assume that Beelink are going to release some sort of proprietary dock, but it would have been nice to have the slot easily accessible through the bottom cover with a standard riser cable.

I’m not sure what the pricing is going to look like as they’re not yet for sale at the time of writing, but I’d imagine they’ll be around $800 for the Ultra 7 series and likely $100 more for the Ultra 9 series.

Let me know what you think of the GTi14 Ultra in the comments section below. What would you use it for?

Raspberry Pi 5 vs Intel N100 PC – Which Is Right For You?

Michael Klements

-

June 27, 2024

9

Raspberry Pi 5 vs Intel N100 PC – Which Is Right For You?

On a couple of my YouTube videos since the launch of the Raspberry Pi 5 last year, people have said that for the price of the Pi 5, you should just get an Intel N100 based mini PC instead. Most cite better video encoding and decoding performance, better OS support, more memory & storage options, and additional PCIe lanes as advantages over the Pi 5. So, today we’re going to compare the two and see whether an N100 Mini PC is a better option and what the limitations of each of them are.

If you don’t know what an N100 PC is. It’s a PC, often in a mini PC form factor, that is built around Intel’s Alder Lake N family, and in this case the N100 CPU. For a long time, Rapsberry Pi’s were substantially cheaper than any newly available Intel hardware, but Pi’s have since crept up in price and this series of processors are now cheap and efficient enough to close that gap to the point whether they’re becoming quite comparable.

Here’s my video of the comparison, read on for my write-up;

Parts Used For This Comparison

Beelink Mini S12 Pro N100 Mini PC – Buy Here
Raspberry Pi 5 – Buy Here
Pi 5 Active Cooler – Buy Here
Pi 5 Power Supply – Buy Here
Pimoroni NVMe Base – Buy Here
Lexar NVMe SSD – Buy Here
Main Power Meter – Buy Here

Pi 5 & N100 PC Setups Being Compared

For this comparison, I’m going to be using the following two setups.

The Pi 5 is an 8GB variant and I’m going to be booting it up from a Pimoroni NVMe base with a Lexar 500GB NVMe SSD. I’ve also added an official active cooler and power supply which, along with the NVMe base and storage drive, comes to a total of $160.

The N100 PC I’ve chosen is the Beelink Mini S12 Pro. This was on special for $159 when I bought it, so it was one of the cheapest options available on Amazon at the time. There were two cheaper options for $154 and $155 but I didn’t recognise either of these brands and I’ve used Beelink products before without any issues so I was happy to pay the extra $5.

So pricing between the two is really similar once you’ve added all of the required components to the Pi 5 and with the N100 we’re getting double the RAM and an included enclosure.

In terms of basic specifications, the Pi 5 has a Broadcom BCM2712 SOC which has a 4-core Arm A76 processor running at up to 2.4Ghz. It’s also got a Videocore VII GPU.

The N100 has 4 4-core Alder Lake N processor running only Intel E cores at up to 3.4GHz and integrated UHD graphics.

Both of these computers have DDR4 RAM. The Pi 5 has 8GB running at 4267 MT/s and the N100 PC has 16GB running at a slower 3200 MT/s.

In terms of storage, both have a 500GB NVMe SSD.

Both computers have similar connectivity options – Gigabit Ethernet, two HDMI ports and four USB ports, although two on the Pi are USB 2.0 instead of all four being USB 3.0 like on the N100 PC.

They both have an M.2 port for an NVMe drive but the N100 also has a SATA port for a 2.5” drive and the Pi has a couple of other interfaces like dual 4-lane camera/display transceivers and a 40-pin GPIO header – we’ll discuss this in a bit more towards the end of the comparison.

The Pi 5 has a single PCIe lane that can run at gen. 3 speeds, to which the NVMe drive is connected. The N100 PC has a built-in M.2 port which makes use of 2 PCIe lanes also running at gen. 3 speeds. So we’d expect the storage speed on the N100 PC to be quite a lot faster than the Pi.

Perhaps the most significant difference between the two is that the N100 is an Intel X86-based system while the Pi 5 is an Arm-based system, so you’ve got far more options for compatible operating systems on the N100 PC than on the Pi 5.

To make testing fair, we’ll be running Ubuntu on both since Ubuntu Desktop 24.04 is available as an officially supported OS through Raspberry Pi Imager and is available for the N100 mini PC as well.

Testing The Pi 5 & N100 PC

To compare the performance of the two, we’re going to run the series of tests below. These should give us a pretty good idea of the capabilities and limitations of each system.

Video Playback at 1080P in a Browser
A Sysbench CPU Benchmark
An NVMe Storage Speed Benchmark
GLMark2 GPU Benchmark
Power Consumption Test

Video Playback at 1080P

Let’s start with video playback at 1080P.

The Pi 5 struggled with this more than I expected it to. It stuttered badly and dropped a significant number of frames at the beginning. Even once playback settled, it still continued to drop frames.

From my experience, the Pi 5 handles video playback in Raspberry Pi OS, which is based on Debian, without any issues, so this is most likely a software issue.

The N100 PC had no problem playing back the 1080P video. Playback was smooth right from the start and was unaffected when running in the window or fullscreen.

So both can handle 1080P video playback but the N100 PC is much better at it.

Sysbench CPU Benchmark

Next, let’s run a Sysbench CPU benchmark. I ran three tests on each computer and then averaged the scores.

I ran the following test on each of the computers;

sysbench --num-threads=4 --test=cpu --cpu-max-prime=20000 --validate run

The Pi 5 managed an average score of 40,359

Actual scores – 40907, 40023, 40148

The N100 PC managed an average score of 44,058

Actual scores – 44022, 44096, 44056

So the N100 PC was about 9% faster than the Pi 5. This is not as significant as I was expecting given the much higher clock speed on the N100’s cores, but there is a small CPU performance gap between the two.

The N100’s results were also far more consistent than the Pi 5, which may suggest that the Pi encounters some sort of thermal limitations when running the tests in quick succession.

NVMe Storage Speed Benchmark

To test the NVMe storage speed, I used James Chamber’s Pi Benchmarks script. This script favours random read/write performance, so is a good representation of how an operating system would make use of the drive.

To run the test, enter the following command in the terminal;

sudo curl https://raw.githubusercontent.com/TheRemote/PiBenchmarks/master/Storage.sh | sudo bash

Over three tests, the Pi 5 managed an average score of 32,089 with average sequential read speeds of 423MB/s and average sequential write speeds of 241MB/s.

Actual scores 31154, 32431, 32683
Actual read speeds 425, 432, 412
Actual write speeds 247, 239, 237

The N100 PC managed an average score of 44,803, so significantly higher than the Pi 5, with an almost 40% improvement. Average sequential reads were around 673MB/s and average writes 495MB/s.

Actual scores 45149, 44992, 44267
Actual read speeds 668, 669, 683
Actual write speeds 507, 491, 487

GLmark2 GPU Benchmark

The N100 has a much more powerful GPU, so I expect it to do a lot better than the Pi in our GLMark2 GPU benchmark.

This benchmark needs to be downloaded and built from source code, and is then run by entering the below command in the terminal;

glmark2

The Pi 5 managed a score of 307.

The N100 PC managed a score of 2070.

So the N100 is over 6 and a half times faster than the Pi 5 in GLMark2, which is obviously a substantial difference.

Power Consumption

Lastly, let’s look at power consumption. This is where I have high hopes for the Pi to stand out.

At idle the Pi 5 uses around 3-4W, and this goes up to 8-9W under load.

The N100 PC uses quite a bit more power, using 8W idle on the desktop and up to 27W under load.

While neither of these figures are particularly high, it’s worth noting that the N100 uses nearly four times the power of the Pi 5. This probably makes little difference on mains but for battery-powered projects that are required to run for many hours or even a few days, the difference can lead to substantial savings in power supply hardware and batteries. This is not all that surprising – ARM computers are known to be power efficient, which is one of the reasons they’re so popular for mobile devices.

Conclusion & Final Thoughts

So, the N100 PC beats the Pi in almost every performance benchmark and comes in at a similar cost.

Test Results For Pi 5 and N100 Comparison

One of the main reasons that people list for getting a Pi over an N100 PC is the GPIO pins, and these are without question much easier to use on the Pi. The GPIO pins are literally available right on the board and there is a wealth of software and tutorials available to utilise them.

That doesn’t mean that you’re out of options for the N100 PC. Microcontrollers like an Arduino Pro Mini or Nano, or even one of these purpose-built Adafruit FT232H USB to GPIO breakout boards make it equally possible to connect tiny OLED displays, read in information from sensors or just work through an introductory flashing LED tutorial on a PC, with relative ease.

This is not as integrated as on the Pi and comes at an additional cost, but for a few dollars might be worth it if you’re just getting started tinkering with electronics.

So, if you plan on using the computer for automation or robotics with a reliance on the GPIO pins then the Pi 5 is the better option, but for experimenting with home server projects, running anything reliant on a GPU, or getting started with Docker or Kubernetes then the N100 mini PC is a great alternative.

I think Raspberry Pi have missed the mark a little with the pricing of the Pi 5. If you are just looking for a cheap computer to get into tinkering with electronics projects then you’re probably better off going for a base version of the Pi 4. This still has plenty of CPU power to run projects locally and you’ll have access to a similar set of IO to the Pi 5 but without the additional cost. After all, a big part of the initial attraction to Raspberry Pi’s was the $35 base price!

Let me know which you prefer and what your use case is in the comments section below.

Mixtile Core 3588E Review

Michael Klements

-

June 6, 2024

0

Today we’re taking a look at the Mixtile Core 3588E. This is a new system on a module, based on the Rockchip RK3588. It’s in the same 69.6 x 45mm form factor as the NVIDIA Jetson TX2 NX module and uses the same 260-pin edge connector – so is compatible with many of the same carrier boards.

Mixtile-Core-3588E-Compared-To-NVIDEA-Jetson

Here’s my video review and testing of the Core 3588E, read on for my written review and results:

Where To Buy The Mixtile Core 3588E

MixtileCore 3588E – Buy Here
A206 Carrier Board – Buy Here

Equipment Used

Infiray P2 Pro Thermal Camera – Buy Here
Video Capture Box – Buy Here

Taking A Look At The SOM & Carrier

In the centre of the module, we’ve got the Rockchip RK3588 processor. This is an 8-core, 64-bit processor that consists of a 4-core Cortex A76 processor running at 2.4GHz and a 4-core Cortex A55 processor running at 1.8GHz. In addition to this, it’s got an Arm Mali G610 GPU.

Alongside it is the eMMC storage module and on the other side of the CPU are the LPDDR4 RAM chips.

The Core 3588E comes in three configurations;

4GB of RAM and 32GB of storage selling for $132
16GB of RAM and 128GB of storage selling for $190
32GB of RAM and 256GB of storage selling for $329

This is quite pricey for a module with this processor on it, given that you can buy a full SBC with ports broken out for this price. But they are fairly close to the pricing on the Turing RK1 modules with the same SOC, so let’s see how it performs.

On the bottom is the 260-pin SO-DIMM connector which allows it to be plugged into devices and carrier boards.

The carrier board that you use will obviously determine which ports and interfaces are available, but the 3588E supports the following basic IO;

HDMI and display port interfaces up to 8K60
USB3.0
USB2.0
PCIe 3.0
PCIe 2.1
UART, SPI, I2C, CAN, I2S, PWM and
Digital IO pins

The carrier board that I’m going to be using with the Core 3588E is the A206, which is designed for NVIDIA Jetson modules.

The main IO is all brought out on one side, with a power input on the left that supports a 9 to 19V DC supply. Alongside it, we’ve got a display port and HDMI port, 4 x USB 3.0 ports, and a Gigabit Ethernet port. The microUSB port at the end is for reflashing the boot loader.

On the top of the carrier board, it has a set of GPIO pins along the right side, a set of pins for buttons and the CAN interface at the back and two camera inputs on the left.

On the bottom is a M.2 E key port as well as an M.2 M key port, an RTC battery holder and a microSD card slot.

Also available for the Core 3588E is an optional heatsink with a built-in PWM fan which plugs into the carrier board. The heatsink is attached directly to the Core 3588e with some small screws that go into thread inserts on the SOM. They provide thermal paste to apply between the CPU and heatsink for improved conductivity.

First Boot & Testing

The board comes preloaded with a custom Ubuntu desktop image, so it’s ready to run right out of the box. You can also compile your own images for Debian and Android.

I’m going to test this board in a similar way to other SBCs that I’ve tried on this channel. I’ll also show you it running a pre-trained AI model to recognise objects in images as this is primarily what these modules are intended to be used for.

We’ll first test some video playback at 1080P, then try to run a Sysbench benchmark, then run a storage speed test, then the AI object detection model and finally we’ll take a look at power consumption.

After the Mixtile Core 3588e has booted up, if we open up HTOP, we can see we have 8 processor cores and our 16GB of RAM. The processor is currently under very little load, being idle on the desktop.

Video Playback At 1080P

First let’s try playing back a YouTube video in Chromium, which I’m going to do at 1080P. We can open up Chromium, go to YouTube and play Big Buck Bunny. I’ll open up stats for nerds and set the playback resolution to 1080P as well.

Video playback in the window is pretty good. We dropped quite a few frames in the beginning but after that playback settles and is very stable and usable.

Big-Buck-Bunny-1080P-Window-Dropped-Frames

It’s also pretty good running full screen. It again dropped quite a few frames in the beginning and then settled down.

If we open up HTOP, we can see that we’re averaging less than 30% CPU utilisation on the first 4 cores, which is relatively low compared to the other RK3588 boards that I’ve tested.

The optional heat sink and fan do a good job at keeping the Core 3588E cool. After 20 minutes of 1080P video playback on YouTube, the CPU was only at 47 degrees and the heatsink was at 38 degrees.

Sysbench CPU Benchmark

Next let’s do a CPU performance comparison with the Mixtile Blade 3, Rock 5 B and Orange Pi 5 Plus which all run the same RK3588 SOC. We’ll do this by running the Sysbench CPU benchmark.

After 10 seconds we have processed a little under 5400 events per seconds and we get a total score of 54,089. Over three tests we get an average score of 54,083.

For comparison, also over three consecutive tests;

Mixtile Blade 3 managed an average of 54,025
Rock 5 B managed an average of 53,642
Orange Pi 5 Plus managed an average of 53,436

So the results from the Core 3588e are slightly higher than the other boards I’ve tested but this is not a significant improvement. It is likely because we’re running a different OS this test was run on Ubuntu and all of the others were tested on Debian.

eMMC Storage Benchmark

Next, we’ll run James Chamber’s Pi Benchmarks script to test the speed of the onboard eMMC storage. This benchmark favours better random read/write performance because this is a good representation of how the storage or drive would typically be used as an OS drive rather than just reading or writing single large files to it.

On completion of the test, we get a total score of 9,822. The individual test results are also listed in the image below.

James-Chambers-Pi-Benchmarks-EMMc-Benchmark

The results aren’t great, sequential read speeds are around 264MB/s and writes are around 225MB/s. Random reads and writes are 13 times and 5 times slower respectively. A better option would probably be to boot from an NVMe drive on the carrier board, but the eMMC storage is ok for an onboard solution if you aren’t transferring large amounts of data.

AI Object Detection Model

Now let’s try an AI object detection model. This is a pre-trained model that you send an image to and it then analyses the image to see if any objects that it has been trained to identify are present.

Here is a list of the objects that the model can detect (each row being a category);

person
bicycle, car, motorbike, aeroplane, bus, train, truck, boat
traffic light, fire hydrant, stop sign, parking meter, bench
cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe
backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket
bottle, wine glass, cup, fork, knife, spoon, bowl
banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake
chair, sofa, pottedplant, bed, diningtable, toilet, tvmonitor, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair drier, toothbrush

We need to run the below commands to download the code, install the dependencies and build the YOLOV5 demo code;

sudo apt install cmake
git clone https://github.com/rockchip-linux/rknpu2
cd rknpu2/examples/rknn_yolov5_demo/
./build-linux_RK3588.sh

Once installed, the model can be run using the below commands;

pushd /home/"Username"/rknpu2/examples/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/
./rknn_yolov5_demo ./model/RK3588/yolov5s-640-640.rknn "Image Name".jpg

I’ve got five test images prepared. We’ll try to put each of these through the model and it’ll produce an output image that shows any detected objects and the model’s confidence in its classifications.

Image 1 is a photograph of 3 elephants;

The image took 18 ms to process, which is impressively fast. It would be able to process around 55 frames per second at this speed.

And this is the result. It got all three correct with a fairly high level of confidence.

Image 2 is a woman in front of a pedestrian crossing with some traffic in the background.

This too took 18 ms to process and the results are pretty good. There is a lot going on in the background but the main elements in the foreground and centre are all correct. Even some of the more obscure background objects are correct.

Image 3 is a similar traffic image.

This has got most of the main elements correct and even a number of the partially obscured cars are correctly identified.

Image 4 is a basket of vegetables with some oranges in front of them.

It made a few mistakes here. I’m not even sure why – these look nothing like an apple or carrot. The confidence levels are pretty low so it clearly had trouble working through these areas.

The last image is a dinner table.

Again this image is mostly correct, even recognising that the whole image is of a dinner table. The knives on either side have been missed and have jointly been labelled a spoon with the spoons next to them. The fork on the far side it got right despite the low confidence.

Power Consumption

Lastly, let’s take a look at power consumption. To measure the Core 3588E’s power consumption, I used a mains power meter. This indicates that the Core 3588E uses about 4W when idle and this goes up to 9W when loaded.

This is a bit higher than the Blade 3 but it does have an active cooler on it and a few extra circuits on the carrier board as well, so is expected but still quite power efficient.

Final Thoughts On The Mixtile Core 3588E

Overall I think that, similar to the Blade 3, the Core 3588E is quite expensive, especially considering it is a bare module and you’d still need to add a carrier board or have a device to plug it into to use it. They have again used good quality components, so you should get good reliability and life out of it, and the module is similarly priced to some alternatives like the Turing RK1 modules.

With the RK3588 SOC, performance is really good, especially considering its low power consumption. This module is ideal for applications like live object detection or motion tracking on a video feed.

Let me know what you think of the Mixtile Core 3588E in the comments section below and if there is anything else that you’d like me to try on it.

I Built A 4-Bay Raspberry Pi 5 Based NAS

Michael Klements

-

May 15, 2024

42

I Built A 4-Bay Raspberry Pi 5 Based NAS

Last year, I built a Pi-based NAS as cheaply as possible using a Raspberry Pi Zero 2W. It was a great project to learn what a NAS is and how to set one up, but it was obviously limited by the capability of the Zero 2W and the cheap storage hardware that was used. So, today we’ll be building a more functional and powerful NAS using a Raspberry Pi 5.

Here’s my video of the build, read on for the write-up;

What You Need To Build Your Own 4-Bay NAS

Raspberry Pi 5 – Buy Here
Raspberry Pi 5 Active Cooler – Buy Here
Radxa Penta SATA Hat (Pi 5 Kit) – Buy Here
Sandisk 32GB MicroSD Card – Buy Here
12V 3A Power Supply – Buy Here
4 x Crucial 2.5″ BX500 SSDs – Buy Here
Noctua 40mm 5V Fan – Buy Here
M2.5 Button Head Screws – Buy Here
M2.5 Brass Inserts – Buy Here
M3 Button Head Screws – Buy Here
Wavelink 2.5G Ethernet Adaptor (Optional) – Buy Here

Equipment Used

Bambulabs X1C 3D Printer – Buy Here
USB-C Pencil Screwdriver – Buy Here
TS100 Soldering Iron – Buy Here
Brass Insert Tips – Buy Here
Gweike Cloud Laser – Buy Here
Acrylic Bending Tool – Buy Here

Some of the above parts are affiliate links. By purchasing products through the above links, you’ll be supporting this channel, at no additional cost to you.

Hardware Used To Build The NAS

The primary piece of hardware that we’ll be using to build the NAS, apart from the Pi 5, is this new Penta SATA hat from Radxa. This hat allows up to 5 SATA drives to be connected to a Raspberry Pi 5 or Rock 5A via their PCEe port.

It’s got 4 SATA ports on the top, which drives plug into top-down, and one eSATA port on the front. Radxa include a cable to plug a fifth drive into this port with the hat. The spacing between the SATA ports allows for 2.5″ drives to be plugged directly into it, but you can also connect 3.5″ drives to it with some extension cables.

Power is supplied to the hat through either a 12V barrel jack on the side or a standard ATX Molex connector on the top. Additionally, you don’t need a second power supply for your Raspberry Pi 5 – the hat will supply 5V to the Pi through the GPIO pins. That’s a really handy feature!

The Penta SATA Hat has got a couple of other ports on it too, like an expansion port for a fan and OLED display on top and an additional fan port at the bottom.

Before mounting the hat onto the Pi 5, we need to add a cooling solution for the Pi’s CPU. I’m going to use a Pi 5 active cooler.

There is one issue with using this cooler and that’s that the end three fins on the heatsink clash with the barrel jack port on the hat. This seems like a bit of an oversight by Radxa but hopefully, they’ll come up with a solution to correct this in future revisions.

My first thought was to add some 6mm spacers between the hat and Pi so that there is a larger air gap between them. This however isn’t possible without also requiring an adaptor for the GPIO pins to still plug into the hat.

The only easy solutions are to either get rid of the cooler or modify the cooler to fit in underneath the connector. Modifying the cooler can be done relatively easily by removing the last three fins, which you can break off with needle nose pliers.

For storage, we’ll use some Crucial BX500 drives as I think these strike a reasonable balance between cost and quality. We’re going to be bottlenecked by the single PCIe lane shared between the drives so there isn’t much point in getting the fastest drives available, any reasonably good quality 2.5″ SSDs would work for the build.

Lastly, we need a microSD card for the operating system. We’ll use a 32GB Sandisk Ultra card for this. I’ve been using these for my Pi projects for years and have had very few issues with them.

I flashed the microSD card with Raspberry Pi OS Lite using Raspberry Pi Imager. This is the base operating system onto which we’ll be installing the NAS software Open Media Vault or OMV. When flashing the operating system image, you may want to change the name of your NAS and you’ll need to enable SSH so that you can log into the Pi remotely once it has booted up to install OMV onto it.

Radxa include hardware with the hat to secure the drives to each other. These make the drive stack a bit more secure, but I’d like to build the stack into an enclosure to better protect the Pi and Penta SATA hat, and provide some cooling to the drives.

Designing The Enclosure

To design the enclosure, I used Fusion360. I started out with a model of the Rapsberry Pi 5, then added the Radxa SATA hat and drives and then modelled the enclosure around them.

My initial thought was to lay the stack down horizontally like a traditional 4-bay NAS, but the Ethernet port on the Pi, the power port on the side of the hat and the power button and activity LEDs on the opposite side mean that it would be oddly proportioned and difficult to get cables plugged into.

So, I decided to keep the vertical arrangement and rather have the drives plug in to the hat through the top. I designed a tray to hold each drive with a pull tab to make it easier to swap out individual drives if needed.

To cool the drives I’ve included a cutout for a 40mm 5V fan on the side which blows air across the four drives and the air then comes up and out the gaps between the drives at the top of the enclosure.

I also added an LED bar to bring the drive activity lights onto the side of the case as well as a button adaptor to allow the Pi’s power button to be pressed and its activity LED to be visible. I’ve included an optional window on the side of the NAS to look into the case to see the drives. I decided on including options with and without this window in the set of print files as I know most people don’t have the tooling required to make the window up but I think it makes the NAS look quite cool.

The enclosure is split into two halves which screw together around the stack, making it easy to pre-assemble and install.

With the design complete, let’s get the components printed out.

Making The NAS Enclosure

I imported the models into Bambu Slicer and set them up to print the main components out in aluminium-coloured PLA with black text. The button adaptor and LED bar are printed in translucent PLA with black sections between the LEDs to separate them. I also added a black accent to the pull tab on each tray.

Download the 3D Print Files

I then sent them to my 3d printer to print out across four build plates.

While the parts are being printed, let’s make up the side panel. This is laser cut from a piece of 2mm clear acrylic and we then use a bending tool to put the 90-degree bend into it.

The window is 108mm x 83mm and the bend line is placed 45mm from the edge. There are notches in the template below to guide placing the bend.

Pi NAS Window Template Download

We now have all of the components required for the enclosure.

To finish off the 3D-printed parts, we need to add some M2.5 brass inserts for the screws to screw into. I’ve also included an option that doesn’t require these inserts in the print files to make it easier to make up but these inserts make the joints a lot more durable so I’d recommend using them if you plan on taking the enclosure apart more than a couple of times.

We also need to glue the window into place using a few drops of superglue or CA glue in the corners.

Mounting The Components Into The Enclosure

Now we can start mounting the components into the enclosure. Let’s start with the fan, which we can mount onto the side with some M3x12mm button head screws. I’m using a 40mm 5V Noctua fan with a thin dust filter between it and the case.

Before mounting the Pi assembly into the enclosure, we need to add the button adaptor to this corner standoff. It just pushes on around the standoff with a very light interference fit so that it is held in place securely but the button is still able to be pushed.

We also need to plug the FPC cable into the hat and the Pi.

FPC Ribbon Cable Plugged Into Pi and Hat

We can then mount the stack to the bottom half of the enclosure using some M2.5x6mm screws through the base.

The status led bar is mounted to the back of the Radxa hat and is held in place with the Radxa hat’s standoffs.

I was going to power the fan using the port on the Pi or the Radxa hat but Noctua don’t have an adaptor to plug into these, so instead I soldered the included adaptor lead to the 3.3V and GND pins on top of the hat. The fan can then plug into this adaptor for power. I chose 3.3V so that the fan runs a bit quieter since it is not a PWM-controlled fan and will be running continuously.

The top half of the enclosure then screws onto the bottom half using six M2.5x6mm screws, three on each end.

We’re now ready to plug our drives in and get our NAS booted up. Each drive is mounted into a tray using four drive screws provided with the Radxa Hat.

Completed Pi 5 NAS Enclosure

With that, our Pi 5 NAS is complete and ready for its first boot and setup.

Drive Trays Visible Through The Side Window

First Boot & Software Setup

We only need power and a network connection to set up our NAS as we’ll be running it headless – meaning we’ll set it up through another computer. So let’s plug the 12V power supply and Ethernet cable into the NAS and it’ll be ready to boot up.

Once power has been turned on, leave the NAS for a few minutes to boot. It usually takes a bit longer to boot up the first time. We can then try to find it’s IP address. This can be done through your network’s DHCP table by logging into your router or by using a utility like Angry IP Scanner. We’re looking for a device called Pi NAS that has recently joined the network.

We can then SSH into the Pi using its IP address to continue setting it up. I’m using Putty on my Windows PC to do this.

Now we need to copy and run this line from the OMV installation instructions GitHub repository to run a script to install OMV on our Pi:

sudo wget -O - https://github.com/OpenMediaVault-Plugin-Developers/installScript/raw/master/install | sudo bash

The installation script takes about 5 minutes to complete and, if successful, should take you to a screen similar to this telling you that the pi is rebooting and your SSH session will be terminated.

When your Pi has rebooted, there is one more thing we need to do before opening up OMV to set up the software. We need to enable the PCIe port on the Pi as this is disabled by default. None of the connected drives will show up until we edit the config file below;

sudo nano /boot/firmware/config.txt

We need to add the below two lines and then reboot the Pi.

# Enable PCIe Port and Set to Gen 3 Speed
dtparam=pciex1
dtparam=pciex1_gen=3

You should then start seeing the activity lights on the drives light up and they’ll show up in the terminal.

Now that all of the installation and configuration work is done, we can access the OMV workbench through a browser by entering the Pi’s IP address. The default login is admin and openmediavault, which you’ll want to change immediately.

There are loads of good guides on setting up OMV, so I’m not going to go through it here in detail but these are the steps I followed:

Set up my four drives in a RAID 5 configuration to balance storage capacity and redundancy. This gives me a total usable storage capacity of 3GB.
Create a storage volume on the array.
Create a shared folder on the storage volume.
Enable SMB file sharing for Windows.
Create a user account with permission to access the shared folder.

Setting up Drives In RAID 5 Configuration

With that complete, we can map the network drive to our PC and can then start using it.

So now let’s see how good it is.

Testing The NAS’ Speed

Copying a single large video file to the NAS, we get an average speed of about 112MB/s which is about 900Mb/s.

A folder of 4,500 smaller files and directories is obviously a lot slower than the single large file but is comparatively as fast as copying them locally.

Copying the large video file from the NAS, we get a similar average speed of about 110-112MB/s.

This looks like we’re saturating the gigabit Ethernet port on the Pi, so next I tried plugging a 2.5G Ethernet adaptor into one of the USB 3 ports on the front.

This made a significant improvement. I instantly got an average of 260MB/s copying files to the NAS although there were a few dips down to about 120MB/s and spikes a little over 270MB/s , so that’s close to saturating the 2.5G ethernet connection which is a great.

Copying the same large video file from the Pi to my PC, I got a little under 200MB/s.

Given the significant speed increase, this is a worthwhile upgrade for less than $20. It really is a bit disappointing that the Pi 5 doesn’t come with 2.5G Ethernet as this makes a big difference to performance for projects that require a large amount of data to be transferred.

Power Consumption

Power consumption is where this NAS shines, especially with its solid-state storage. At idle, the NAS uses a miniscule 9W and consumption only goes up to around 12 under load. This is significantly less than the 30-40 watts that a typical 4-bay home or small office NAS uses. My Asustor NAS uses about 18w at idle with the drives spun down.

Final Thoughts On My Raspberry Pi 5 Based NAS

So that’s my new 4-bay Pi 5 NAS complete. I’m really impressed with the speeds that I managed to get using the 2.5G Ethernet adaptor. This highlights one of the weaknesses in the Pi 5, which really should have been designed with a 2.5G Ethernet port given its relatively recent release date.

Cost-wise, this is not the most affordable NAS on the market but it’s also not particularly expensive considering it is very versatile and customisable, running open-source hardware.

The main NAS components, being the Pi ($80), Radxa Penta SATA Hat ($45), Cooler ($5), Power Supply ($15), MicroSD Card ($5) and Fan ($15) come to $165 – which is around the lower end of what a commercially available 4-bay NAS would cost, albeit without the DIY work required.

The NAS is really power efficient for those who live in an area or country where power is expensive. Running this NAS 24/7 for a year would cost less than a third of what a traditional NAS would cost to run.

Let me know what you think of it in the comments section below and feel free to send photos through of your build if you 3D print your own.

Is The New Orange Pi 5 Pro A Good Raspberry Pi 5 Alternative?

Michael Klements

-

April 24, 2024

2

Today we’re going to be taking a look at the Orange Pi 5 Pro. This is a new SBC from Orange Pi that is based on the Rockchip RK3588S SOC. The RK3588S is largely similar to the RK3588, with the main difference being a reduction in PCIe lanes from 4 to just 1.

Given the form factor, IO and port layout, the Orange Pi 5 Pro is clearly targeted as an alternative to the Raspberry Pi 5. So let’s see how it’s price, performance and usability stack up.

Here’s my video of the Orange Pi 5 Pro, read on for the written review:

Where To Get The Orange Pi 5 Pro

Orange Pi 5 Pro (Amazon) – Buy Here
Orange Pi 5 Pro (Aliexpress) – Buy Here

Equipment Used

Infiray P2 Pro Thermal Camera – Buy Here
Power Meter USB C Cable – Buy Here
Video Capture Box – Buy Here

Unboxing & First Look At The Orange Pi 5 Pro

The Orange Pi 5 Pro comes in the usual transparent plastic sleeve with an Orange Pi branded sleeve around it. Within the case, the 5 Pro is protected by an anti-static sealed bag. It includes a WiFi antenna but I’m going to remove this for testing as I’ll be using a wired connection.

Taking a look around the board, the RK3588S processor is in the middle. This is an 8-core, 64-bit processor that consists of a 4-core Cortex A76 processor running at 2.4GHz and a 4-core Cortex A55 processor running at 1.8GHz. In addition to this, it’s got an Arm Mali G610 GPU. So we should get really good performance from it with double the CPU cores of the Raspberry Pi 5, although four cores are running at a lower clock speed.

Alongside the processor is the RAM chip. The Orange Pi 5 Pro comes in a 4GB, 8GB and 16GB variant, all with LPDDR5 RAM, which is a step up from the LPDDR4 RAM on the non-pro version of the Orange Pi 5.

Above that is the WiFi 5 and Bluetooth 5 chip, with a DSI port alongside it.

Similar to the Raspberry Pi 5, the Orange Pi 5 Pro, has a Gigabit Ethernet port on one side alongside four USB ports. Notably, three of these are USB 2.0 and only one is USB 3.0, unlike the Raspberry Pi’s two USB 3.0 ports.

They have also included a header for additional USB 2.0 ports behind the physical ports, which could be useful for building the Orange Pi 5 Pro into an enclosure.

Along the other side is an HDMI 2.0 port, an audio port, an HDMI 2.1 port and a USB C power port. The HDMI 2.1 port can do up to 8K 60 and the HDMI 2.0 port up to 4K 60. I like that they’ve made space for full-size HDMI ports, I don’t really like how fragile the micro HDMI ports on the Raspberry Pi 4 and 5 are, often requiring adaptors or special cables to plug into them rather than commonly available full-size HDMI cables. Interestingly, they’ve chosen to keep the audio port since the Raspberry Pi 5 did away with it.

On the opposite side is a 40-pin GPIO header which follows the same pinout as the Raspberry Pi boards, it’s also colour-coded which makes it a bit easier to locate the 5V, 3.3V and ground pins. Next to the pins are a 5V fan connector and an RTC connector.

Underneath the board is a prominent M.2 M-key port which allows you to connect an NVMe SSD to it. This is only a PCIe gen 2 x 1 port, so won’t be that fast by today’s standards. It’s also got a connector for additional eMMC storage, a microSD card slot and two camera ports.

To power the Orange Pi 5 Pro, you can get a 5V 5A USB C power supply from Orange Pi’s Aliexpress store. This has the same specs as the Raspberry Pi 5’s official power supply so you can use one of those too.

So that’s an overview of the hardware. As you can see, the IO is somewhat limited due to the single PCIe lane on the RK3588S, but you’ve still got a range of options and support for an NVMe drive without an adaptor. Hopefully, the power of the CPU and improved RAM will make up for some of the IO limitations. So let’s get it booted up to find out.

Operating System Options For The 5 Pro

Like with other Orange Pi boards, they have a number of operating system images available to run on the 5 Pro, including the usuals like Ubuntu, OpenWRT, Debian and Android. The main OS that they want you to run on it is called Orange Pi OS and they also have a couple of options for this, including an Arch Linux and Android version as well.

At the time of writing this review, not all of the above images are available, but I’ll be using the Debian image for testing as this is what I’ve tested other SBCs on. This allows the results of my other tests to be a bit easier to compare to.

Installing an operating system is really simple. You just download the relevant image from their website and then flash it onto a microSD card using a utility like Balena Etcher.

I’ve got the OS flashed onto a microSD card which the Orange Pi 5 Pro will be booting off and I’ve also added an NVMe storage drive to do a benchmark on as well.

Testing The Orange Pi 5 Pro’s Performance

I’m going to test the Orange Pi 5 Pro in the same way I usually test SBCs, by running the following tests:

Video playback at 1080P
Video playback at 4K
Sysbench CPU benchmark
NVMe drive speed test
Measure power consumption

The first boot on a new operating system takes a little longer than subsequent boots but we’re still on the desktop in under a minute.

If we open up HTOP, we can see we have our 8 processor cores listed. They’re all relatively idle without any applications running, and we can see our 16GB of RAM.

We can also see that our NVMe drive is connected, which is also apparent through the desktop icon.

Video Playback At 1080P and 4K

First, let’s try playing back a YouTube video in the default browser Chromium. I’m going to do this at 1080P and then at 4K.

We’ll set the display resolution to 1080P. Then let’s open up the browser, go to YouTube and open up Big Buck Bunny. I’ll set the playback resolution to 1080P and then open up stats for nerds.

Video playback in the window drops a few frames in the beginning but after that is near perfect.

Playback is ok in full screen, but it does drop a couple of frames every so often which is noticeable and does become annoying. You can see example video clips in my video linked at the beginning of the post.

Now let’s step it up to 4K. There was a bit of weird behaviour when changing to 4K – the display wouldn’t refresh when the setting was applied and you could clearly see that the resolution hadn’t changed, although it indicated that it had. A reboot after changing the resolution setting seemed to fix the issue and it then booted up in 4K.

Now we can reopen the browser and Big Buck Bunny. We’ll set the playback resolution to 4K as well.

In 4K, playback gets off to a pretty poor start. We dropped a large number of frames in the beginning and continue to drop frames throughout playback. Opening up to full screen is even worse, to the point where it is basically unusable.

So this board does a fair bit worse at 4k playback than the Orange Pi 5 Plus. This is probably a software issue with the board using software decoding instead of hardware decoding. Opening up HTOP while the video playing back, the CPU is at about 40-50% utilisation across all cores which is a lot more than the 20-30% CPU utilisation on only 4 cores that we had on the 5 Plus playing back the same video.

To jump in here, if you’re going to be primarily using the Orange Pi 5 Pro as a media device then you’ll likely want to go with the Android image as this tends to perform better for video playback.

Sysbench CPU Benchmark

Next, let’s check the performance of the 5 Pro by running the Sysbench CPU benchmark.

After 10 seconds we have processed a little under 5,350 events per seconds and we get a total score of 53,519. Over three tests, the Orange Pi 5 Pro managed an average of 53,520.

For comparison, and also over three consecutive tests;

The Rock 5 B managed an average of 53,642
The Khadas Edge 2 managed an average of 51,568
The Orange Pi 5 Plus managed an average of 53,436
The Raspberry Pi 5 manages an average of 35,000

So the results of the Orange Pi 5 Pro are a bit better than the Khadas Edge 2 with the same SOC. They’re actually very similar to the Rock 5 B and Orange Pi 5 Plus which have the better RK3588 CPU and they’re significantly higher than the Raspberry Pi 5, with an over 50% improvement.

Thermally, you’ll probably need to use a heatsink on the CPU if you’re running CPU-intensive tasks. After playing back 4K video for around 15 minutes, the Orange Pi 5 Pro’s CPU was at over 70°C and the surface of the CPU was at 55°C. This was in a relatively cool room with an ambient temperature 20°C.

NVMe Drive Speed

Next, let’s look at the NVME drive speed. For this test, I’m going to be using James Chambers PiBenchmarks script. I’ve used this script recently to benchmark a few different NVMe hats for the Raspberry Pi 5.

James Chambers PiBenchmarks Script Running

I ran the test three times, with very consistent results across the three tests and an average total score of a little over 16,000. The Pi 5 had results slightly under 60,000, so the Orange Pi 5 Pro is significantly slower than the Raspberry Pi 5, which is likely mainly due to the one-generation reduction on the PCIe port.

Power Consumption

Lastly, let’s take a look at power consumption. To measure the 5 Pro’s power consumption, I used a USB power meter cable that supports power delivery.

This showed that the Orange Pi 5 Pro uses about 3w when idle and this goes up to 6-8w when loaded, with peaks of up to 9w. So fairly similar to the Raspberry Pi 5 and a bit more power-hungry than the Khadas Edge 2 with the same SOC.

Final Thoughts On The Orange Pi 5 Pro

In terms of cost, only the 16GB variant of the Pi 5 Pro is currently available at the time of writing this review, and it retails for $109. This is about $30 more than the best variant of Raspberry Pi, the 8GB version, but you’re getting double the RAM, double the CPU cores and onboard NVMe support which means you won’t need to buy an additional hat for it.

So, I think that the price is fair in terms of value for money.

Overall I think that the Orange Pi 5 Pro is a great alternative to the Raspberry Pi 5 if your project favours raw CPU or GPU performance – so for computationally intensive projects and simulations. It offers much better raw performance than the Raspberry Pi 5 and has a decent set of built-in IO.

It is limited by the single PCIe lane, so you don’t get two USB 3 ports, and the NVME drive is only running at gen 2 speeds. Depending on your project, this is might be something that you can live with.

The main reason why you’d want to get a Raspberry Pi 5 over the Orange Pi 5 Pro is for the software support. Orange Pi is one of the best-supported alternate SBC manufacturers, but even so, their community software support is quite far behind that of the Raspberry Pi 5. Raspberry Pi have fostered a large community around their products and this community is really good at working together to develop software and troubleshoot issues.

The Orange Pi 5 Pro is a great board if there is a software package or OS image built for it or if you have a good understanding of software and programming, but for the average hobbyist or tinkerer, if you’re trying a project that hasn’t been tried before, or you’re running a project that utilises the GPIO pins then you’re probably better off with the Raspberry Pi 5.

Let me know what you think of the Orange Pi 5 Pro in the comments section below and if you’ve got anything else that you’d like me to try run or test on it.

Which NVMe Hat Is The Best For A Raspberry Pi 5

Michael Klements

-

March 27, 2024

6

Which NVMe Hat Is The Best For A Raspberry Pi 5

If you don’t know already, I’ve been selling these 3D printed cases for Raspberry Pi’s online for a few years now. With the launch of a number of NVMe drive hats for the Pi 5’s PCIe port, I get asked quite a lot which hat is best for it and which case to choose. So today, I set out to find out.

Here’s my video of the test, read on for the write-up and detailed results:

These NVMe hat’s have a few common features, so let’s have a look at those first.

They all connect to the Pi 5 through an impedance-controlled ribbon cable at the back of the Pi and then either sit on top of or underneath it. They feature an M.2 M-Key port that the drive plugs into and although the Pi supplies power to them directly through the ribbon cable, they often have an option for an external power source as well.

To accommodate these hats, I’ve got two case designs. One which takes the Pineberry HatDrive! Top and another which takes either the Pimoroni NVMe Base or the Pineberry HatDrive! Bottom.

Purchase Links For Components Used In This Test

NVMe Hats Tested:

Pineberry HatDrive! Top – Buy Here
Pineberry HatDrive! Bottom – Buy Here
Pimoroni NVMe Base – Buy Here
Geekworm X1001 – Buy Here

Test Components & Equipment Used:

Raspberry Pi 5 – Buy Here
Sabrent Rocket 4.0 – Buy Here
Pi 5 Power Supply – Buy Here
Pi 5 Active Cooler – Buy Here
ClonerAlliance UHD Pro Video Capture – Buy Here

Is A Top or Bottom Mount Hat Better?

In terms of which physical layout is best, I have a preference for the top-mounted hat but there are a lot of benefits to the bottom-mount as well.

The top mount allows you to fit a Pi 5 active cooler in between the hat and the Pi, so that takes care of cooling, and you’ve then got the hat and NVMe drive on the top. This leaves you plenty of room to add a heatsink to the NVMe drive and it stays reasonably cool without a heatsink just because it isn’t boxed in underneath another board. Its also quick and easy to swap the drive out for a different one if you’re switching operating systems or storage.

The drawbacks of the top-mounted NVMe drive are that the Pineberry version is limited to a more compact 2230 and 2242 size drive. These are a little bit less common and more expensive. You also don’t have access to the Pi’s CPU to put a larger cooler like an Ice Tower onto it and it blocks some of the GPIO pins.

NVMe Drive Accessible Through The Top Of The Stack

The bottom mount has the NVMe drive underneath the Pi which means you can now access all of the GPIO pins and add a larger cooler onto the CPU. You can now also use 2280 size drives, and in the case of the Pimoroni NVMe base, 2260 drives as well.

The drawbacks of the bottom mount are that the NVMe drive is covered and is in a relatively small space, so it gets hot. You’re also quite limited in options for a heatsink since it has to be very compact. As someone who experiments quite a bit with different operating systems, I find having to disassemble the stack to get to the drive the biggest drawback and the main reason why I prefer the top mounted hat.

So that’s an overview of the physical differences, pros and cons. I’d say that if you tend to need to swap NVMe drives around often then you’d probably prefer the top-mounted hat but if you’re happy to install a drive and leave it in place long-term then the bottom mount is probably the more versatile option.

Performance Testing The NVMe Drive Hats

I’m going to be testing three different NVME drive hats.

First up is the Pineberry HatDrive. The HatDrive Top and Bottom have the same onboard components and circuitry, just a different layout, so I’ll use the HatDrive Top for testing and the results as a representation of both.

Next is the Pimoroni NVMe Base. This offers a wider range of drive size options than the HatDrive options but only comes in a bottom mount variant.

Lastly, we’ve got the Geekworm X1001 NVMe shield. You don’t get any additional PCB for your money with this board, they’re really kept it as compact as possible. Similar to the Pimoroni base, it supports four different size drives, but is a top-mount hat.

In terms of cost, from the manufacturer’s official websites, converted to US dollars and excluding shipping;

The HatDrive Top costs $21 and the bottom variant is a bit more, costing $24.
The Pimoroni NVMe base is a bit cheaper at only $14.
The Geekworm X1001 is a dollar more than the Pimoroni hat at $15.

So Pineberry’s boards are a fair bit more expensive than the other two.

Next, let’s take a look at the performance of each hat.

For this, I’m going to use the same Raspberry Pi 5 with an active cooler installed, and the same NVMe drive for each test which I’ll swap between the hats.

I’m using a Sabrent Rocket 4.0 as this drive is listed as officially supported on all of the hat’s product pages. It’s also known to be a reliable and fast drive. It is probably a little overkill as it’s a Gen 4 drive and the Pi only supports up to Gen 3, but at least we’ll know that the drive isn’t the bottleneck. I’m using a 2230 size drive so that it is compatible with all of the hats since the Pineberry HatDrive! Top only supports 2230 and 2242 size drives.

To test performance, I’m going to be using James Chamber’s PiBenchmarks script. This benchmark favours better random read/write performance, but this is a much better representation of how the drive would typically be used as a OS drive rather than just reading or writing single large files to it. This benchmark will run on SBCs running most Linux distributions, so you can try it out on your setup as well.

As I mentioned earlier, the Pi only supports PCIe Gen 3, but this is not supported by default so we’ll need to modify the Pi’s config file to enable it. We just need to add the below line to the config file [/boot/firmware/config.txt] and then reboot the Pi.

dtparam=pciex1_gen=3

Let’s start testing with the Pineberry HatDrive! Top, which I’ve now got connected up.

With the Pi rebooted, we can obviously see our NVMe drive.

Running the benchmark is as simple as copying the below single line into your terminal and hitting enter.

sudo curl https://raw.githubusercontent.com/TheRemote/PiBenchmarks/master/Storage.sh | sudo bash

Terminal Line To Run PiBenchmarks Script

I ran the test three times and got the following average results with an average total score of 60,011.

Next up is the Pimoroni NVMe base.

Running the same script three times, I got the following average results with an average total score of a fractionally lower 59,875.

Lastly, I tested the Geekworm X1001 hat.

I got the following average results with an average total score of 59,950.

You can download a full table of the results here:

NVMe Hat Test Results Download

Looking at the combined results, they all performed quite similarly, with almost all results being within 1% of each other and most within 0.5%. The only result that was outside of this was the Disk Write speed, which was within 3%. This had the Pineberry HatDrive! performing the fastest and the Geekworm X1001 performing the slowest.

Final Thoughts On The NVMe Drive Hats

The similar results are not all that surprising, since the NVMe controller is physically located on the NVMe drive, which we’re swapping between hats. These hats with a single M.2 port are actually quite simple and most of the onboard components are for the power to the drive and status LED circuitry. There could have been design issues like incorrect impedance matching that may have affected the results, so it was worthwhile doing the test to demonstrate that we’re getting similar results from each of them.

I guess the takeaways from the results are that the most significant considerations in deciding between the NVMe drive hats are the cost and whether to go with a top or bottom hat.

From my hands-on experience with all three of these hats, the Pineberry Hat and the Pimoroni Hat seem to be better quality than the Geekworm one. The Pimoroni one is the best value for money, so go with that one if you are happy with a bottom mount hat.

Pineberry and Pimoroni Options Are Better Quality

If you want a top mount hat then you’ll need to decide whether you value the lower price of the Geekworm one or favour the slightly better performance and quality of the Pineberry hat.

Pineberry and X1001 Options For Top Mount Hat

Let me know which hat you prefer or if there are some other drives you’d like to see me test with them in the comments section below.

Track Aircraft In Real-time With Your Raspberry Pi Using The FlightAware Pro

Projects

Michael Klements

-

March 13, 2024

0

Track Aircraft In Real-time With Your Raspberry Pi Using The FlightAware Pro

Have you ever seen a flight overhead and wondered where it is going? Or seen a unique-looking aircraft and wondered what type or model it was? Well, today I’ve got something exciting to share with the aviation enthusiasts out there. We’re going to set up our own flight tracker using the FlightAware Pro USB stick and a Raspberry Pi. This is a really easy and fun project that allows you to track aircraft in your area in real-time.

Here’s my video of the project, read on for the write-up;

What You Need To Build Your Own FlightAware Tracker

Raspberry Pi 4 – Buy Here
FlightAware Pro – Buy Here
Indoor Antenna – Buy Here
MicroSD Card – Buy Here
Pi 4 Power Supply – Buy Here
Pi 4 Basic Heatsink – Buy Here

What Is A FlightAware Tracker?

First things first, let’s talk about the FlightAware Pro USB stick. This little device is a dedicated ADS-B, or Automatic Dependent Surveillance-Broadcast, receiver that simply plugs into a free USB port on your Pi.

ADS-B is a technology that enables aircraft to determine their position via satellite and then broadcast it. This information can be received by ground stations, like the one we’re going to be building, and this enables aircraft to be tracked. Using an antenna with the FlightAware USB stick, you can track aircraft up to 400km or 250 miles away.

Before we jump into the assembly and setup process, let’s take a look at what FlightAware actually is. FlightAware is an online platform that provides real-time flight tracking, and not just the flight information like departure and estimated arrival time that you can find on Google, this gives you full flight tracking of airspeed and altitude, flight paths, aircraft information and historical data.

You’re also not limited to only commercial aircraft, you can track commercial and private traffic and even the occasional military aircraft. By creating your own tracker, which is called a PiAware tracker, you’re contributing to FlightAware’s network of over 30,000 ground stations, enhancing the accuracy of global flight tracking.

In exchange for this, they provide you with a free Enterprise user account which gives you full access to their platform, so you can see flights that are out of range of your receiver as well!

Assembling Your PiAware Tracker

To build your own FlightAware flight tracker you’ll need to add a Raspberry Pi, microSD card and power supply to the FlightAware USB stick and antenna.

Next, let’s dive into setting one up, and there really isn’t a whole lot to it. First, we need to flash the operating system to our microSD card. This is done by downloading the prepared OS image from the FlightAware website and then burning it to the microSD card using an imaging utility like Etcher.

The image is ready to run so you don’t need to do anything else if you’re using a wired Ethernet connection like I’m going to. If you want to use WiFi then you’ll need to follow their configuration steps (under item 3) to add your WiFi network’s information to the card so that your Pi knows how to connect to it.

Once your card has been flashed, plug it into your Pi’s microSD card slot.

To complete the assembly, we need to plug the components into the Pi. The Flight Aware USB stick goes into one of the USB ports – you’ll need to use the USB 2.0 ports as the USB 3.0 ports are too close to the Ethernet port to allow a cable to plug in next to it.

The antennae plugs into the USB stick and a little retaining nut locks it into place.

Then add the Ethernet cable and plug in the power supply to boot it up.

First Boot & Associating It With Your Account

It’ll take a couple of minutes for the first boot and while that’s happening, let’s head over to FlightAware’s website to sign up for a free account.

Once that is done we need to find our tracker’s IP address on our local network. There are a few ways to do this, you can use a utility like Angry IP Scanner or your can look at your network’s DHCP table. I’ve looked in my DHCP table and found my PiAware device and IP address.

We can then enter the IP address into a browser on the same network to access the Pi. From here you’ll be asked to associate your PiAware tracker with your account by logging into it. Once you’ve done that, you’re now officially part of the FlightAware network and you’ll see your account has been upgraded to an enterprise account.

Your Pi will then immediately start contributing data to the FlightAware network.

What You Can See Through The FlightAware Web Interface

I’ve had my Pi running now for a little under three months. From the FlightAware site, you can see your feeder’s status and when last information was received from it. You can also see stats and graphs for the number of aircraft reported throughout the past 24 hours, which direction they were reported from and even the type of aircraft positions received.

They also give you a ranking that ranks your FlightAware feeder against other contributors. I’m still not too sure how they arrive at the total rank position but they give you some stats on your reported positions relative to others. Obviously, if you’re in a busier airspace then you’re going to be contributing far more to the network than others, but the network relies on having contributors in remote locations too.

You can also see the flights that have been tracked by your feeder in the last hour and the sites around you, including when last they reported positions.

The really exciting stuff is on the SkyAware Anywhere page. Here you can watch flights around the world in real-time. This is the area around Sydney where my feeder is based and the aircraft are coloured according to their altitude.

You can see a summary of those in view on the right but you can also click on aircraft to get more detailed information on them and show their flight path since you opened the window. We can also visit an aircraft’s flight page to show even more information about it.

At the top of the page is the flight information that you’d typically see if you did a Google search for the particular flight number.

Below that is a map showing the planned flight path and the planned and actual altitude and airspeed.

The flight that I chose has only recently taken off, flying to Hong Kong. Let’s pick a different flight that looks like it is coming in to land and we’ll be able to see the altitude and airspeed history for the flight.

The next one that I chose is a regional domestic flight by REX from Coffs Harbour to Sydney. Now we can see the actual flight path taken alongside the plan and also the altitude and airspeed history – we can see that it is in its descent phase and coming in to land soon.

For each aircraft, you can also see its planned upcoming flights, the current flight in progress and a log of past flights. You can click on any of these to see the flight plan and/or actual flight data.

On the right-hand side are the aircraft details like the owner and operator, registration details and the current flight data. They also give you a gallery showing images of that aircraft type.

So on the top right is a similar aircraft by the same carrier but with a different tail number.

FlightAware doesn’t only track commercial aircraft, we can also see small private aircraft. Here is one doing circuits around a small airport, so it’s probably a flight training exercise.

You can also go into an aircraft’s flight history and have a look at the details from any past flights. This will show the flight path taken, airspeed and altitude and you can even open up a full log and see all of the position reports and which facility reported them.

There is also a mobile app for iOS and Android that offers the same information as the web page but in a mobile-friendly format. I actually prefer using the app and find it a bit more intuitive.

Final Thoughts On The FlightAware Tracker

So this is an exciting way to dive into the world of aviation tracking and contribute to a global network of aviation enthusiasts. It is a relatively inexpensive project that can even be set up on a Pi Zero W or Zero 2 W to save costs and you get a lot out of it, especially if you have an interest in aircraft.

Mine has been running flawlessly for 3-4 months now and I’ve really enjoyed routinely hopping onto the app when I see an aeroplane or helicopter overhead and wonder what it is doing or where it is going. It’s also amazing to have access to detailed flight logs for any flight or aircraft number.

Let me know what you think of the FlightAware tracker in the comments section below and if you’ve got any questions on what it can and can’t do.

Pi 5 Desktop Case For NVMe Base or HatDrive! Bottom

Michael Klements

-

March 8, 2024

4

Pi 5 Desktop Case For NVMe Base or HatDrive! Bottom

Today we’re going to be assembling a 3D-printed case for the Raspberry Pi 5 and Pimoroni’s NVMe Base or Pineberry’s HatDrive! This is an adaptation of my standard desktop case for the Raspberry Pi 5, with adjustments made for the bottom-mounted NVMe hat. This case also includes an adaptor so that you can still use the power button on the back of the Pi 5.

Pi 5 Desktop Case For NVMe Bottom Side 2

What You Need To Assemble Your Case

Raspberry Pi 5 – Buy Here
MicroSD Card – Buy Here
Pi 5 Power Supply – Buy Here
Case Kit (3D Printed Parts & Screws) either of the below:
- Pimoroni NVMe Base Case Kit – Buy Here
- HatDrive! Bottom Case Kit – Buy Here
NVMe Hat, either of the below:
- Pimoroni NVMe Base – Buy Here
- HatDrive! Bottom – Buy Here
Ice Tower Cooler (for the Pi 5) – Buy Here
2280 NVMe Drive – Buy Here
Or Alternate Cooling Solution Below
Raspberry Pi 5 Active Cooler – Buy Here
40mm 5V Fan – Buy Here

Ensure that you get the Ice Tower cooler for the Raspberry Pi 5, the cooler for the Pi 4 will not fit.

Assembling The Raspberry Pi 5 NVMe Base Case

To start, we need to install the M2.5x6mm brass standoffs that come with the case kit on the bottom of the case. These are installed with the male thread facing upwards into the case and are each held in place with an M2.5 button head screw through the base of the case.

Install M2.5 Standoffs Into Bottom Of Case

Secure Standoffs With M2.5 Button Head Screw

Next, install your NVME SSD onto your hat. I’m using a Crucial drive on mine. Both Pimoroni and Pineberry also have a list of compatible drives that have been tested on their website. Plug your PCIe ribbon cable into the socket on the hat as well – make sure that the orientation is correct.

Place the assembled shield onto the brass standoffs and then use the M2.5x7mm brass standoffs supplied with your Ice Tower cooler to hold it in place. Don’t install an Ice Tower standoff on the hole nearest to the power port, this is where the button adaptor will go. Instead, install one of the female-to-female standoffs from your hat kit. If you are using the Pineberry HatDrive! then you’ll need to install a small 1mm black spacer from the case kit underneath this standoff as it is 1mm shorter than the Pimoroni ones.

Install NVMe Base and NVMe Drive - With Ice Tower

If you are not using an Ice Tower cooler with your hat, then install four female-to-female standoffs to hold the hat in place. You’ll then use the short screws included with your hat kit to hold the Pi in place on the hat. If you’re using the Pineberry Hat, you’ll need to use the four 1mm black spacers supplied with the case kit underneath the hat as the standoffs are slightly shorter than the ones supplied with the Pimoroni kit.

Install NVMe Base and NVMe Drive - No Ice Tower

Next place the Raspberry Pi on top of the hat and secure it with three of the M2.5x7mm brass standoffs that came with the Ice Tower cooler. Don’t install a standoff on the hole nearest to the power port, this is where the button adaptor will go.

Plug the PCIe cable into the Pi. The tab on the connector can be pushed down to secure the cable through the microSD card slot. Likewise, to release it, the tab can be pushed up through the slot as well.

Next, we can position the button adaptor over the remaining mount. Slide the button adaptor into position through the microSD card slot at the back of the case in the orientation shown below. It should go in easily – do not force it past the LED or button as you may damage them. If you feel resistance, rather try to remove and reposition it until it slides into place without interfering with the surrounding components.

Now we can install our Ice Tower cooler. Before we install it, we need to remove the fan by removing the four M3 screws in the corners. We’re going to be installing the fan onto the side panel.

Remember to add the thermal pad to the CPU before putting the cooler into place. Secure the cooler with three M2.5 thumb screws, one into each of the brass standoffs. Again leave the button adaptor unsecured for this step.

Install Ice Tower and Secure with Standoffs

Now use the M2.5x12mm button head screw that was supplied with the case kit to hold the button adaptor in place. You’ll need to add the thicker black spacer supplied with the kit between the button adaptor and the leg of the Ice Tower cooler as well.

Do not overtighten the screw as you need the button adaptor to be able to still move slightly to push and release the button. You should be able to feel the button press and release easily through the button adaptor.

To install the fan, we’re going to use the same method that I used on my other case designs where the screws do not go all the way through to the back of the fan. We instead press the M3 nuts into the front of the fan and the screws then hold these in place against the side panel.

It is easiest to press these into place by placing the nuts onto a flat surface and then pushing each pocket in the corner of the fan down onto the nut. The nut is in position when it is flush with the face of the fan.

Screw the fan to the side panel using four of the included M3x8mm button head screws.

Then plug the fan into your Raspberry Pi’s fan port or GPIO pins (if you’re using your own fan) before screwing the side panel into place.

Close up the case with the two side panels and four M3x8mm button head screws on each side.

Your case is now complete and ready to run. Follow Pineberry’s documentation or Pimoroni’s documentation for instructions on booting your Pi from the NVME drive.

Is It Worth Water Cooling A Raspberry Pi 5?

Michael Klements

-

February 7, 2024

4

Is It Worth Water Cooling A Raspberry Pi 5?

Today we’re going to be taking a look at a new water-cooling kit that has been designed for a Raspberry Pi 5. We’ll see how its custom water block and performance stack up against the DIY solution that I built for my Pi 4 a few years ago. I also used this solution on my Pi cluster, cooling 8 Raspberry Pi 4s at once. It worked really well for that too, so I expect it’ll also work well on the more powerful Pi 5.

Here’s my video of the comparison, read on for my write-up:

Purchase Links For Parts & Equipment Used

Raspberry Pi 5 – Buy Here
MicroSD Card – Buy Here
Pi 5 Power Supply – Buy Here
Pi 5 Active Cooler – Buy Here
DIY Water Cooling Block Kit:
- Green – Buy Here
- Red – Buy Here
- Blue – Buy Here
Water Cooling Kit – Buy Here

Tool & Equipment Used:

Gweike Cloud CO2 Laser Cutter – Buy Here
InfiRay P2 Pro Thermal Camera – Buy Here
Jobi Gorilla Pod – Buy Here

Unboxing The Seeed Studio Water Cooling Kit

The water cooling kit that we’re going to be using has just been launched, it is available from Seeed Studio’s web store and currently costs $120.

It includes a fan and water pump assembly, a cooling block, a 12V power adaptor, the installation hardware like some thermal pads and screws as well as some flexible tubing to run to the water block.

The water block is a custom design for the Pi 5 and the kit comes with a black aluminium heatsink for the bottom of the Pi as well.

Outlining The Thermal Testing Process

To test the thermal performance of the manufactured water cooling option, we need something to compare it to. For that, I’m going to try two different cooling solutions. The first is the active cooler that is designed for the Pi 5. This is a cheap and simple solution, costing just $5, and is also commonly available. It’s got an aluminium heatsink covering the CPU and surrounding heat-generating components with a small PWM fan on it.

The second is my original water cooling setup from my Pi 4. I’ve had to make some changes to it to work with the Pi 5 – I’ve redesigned the heatsink bracket for the new CPU location and I’ve strengthed the legs and holder for the Pi as these were a little flimsy on the original. Other than that it’s pretty much the same set of components, especially in the cooling loop.

I’m going to use the same Raspberry Pi 5 with each cooling solution for consistency and I’ll just switch it between each solution for the test. I’m also going to be using the official Pi 5 power supply so that we don’t run into any power or undervoltage issues.

To test the thermal performance of each solution, I’m going to use a utility that I used to test my previous setup called CPU burn. It can be installed on your Pi using the following commands:

wget https://raw.githubusercontent.com/ssvb/cpuburn-arm/master/cpuburn-a53.S

gcc -o cpuburn-a53 cpuburn-a53.S

It is then run in the terminal with the below command, which will also display the CPU temperature and clock frequency:

while true; do vcgencmd measure_clock arm; vcgencmd measure_temp; sleep 10; done& ./cpuburn-a53

This utility fully loads each of the CPU cores. I’ll leave this running for about 10 minutes on each so that we reach the point of equilibrium every time. I expect both of the water cooling setups to perform quite well, so we’ll also try overclocking the Pi to 2.8GHz and see how each handles the additional overclocking heat as well.

I did try to get my Pi overclocked to 3Ghz but I couldn’t get it to be stable enough to survive a 10-minute stress test with CPU Burn at this CPU frequency.

I’m going to set up a thermal camera to watch each solution during the test so that we can see any hot spots. The thermal camera doesn’t work on metallic surfaces, so it will be a bit limited but we should still be able to see any significant issues.

Just out of interest, I ran the test first without any cooling solution connected and ran into thermal throttling after about half a minute. Thats evident by the drop in CPU clock frequency once we hit 85°C.

Thermal Testing The Three Cooling Solutions

Testing The Raspberry Pi 5 Active Cooler

I’m going to start with the Pi 5 Active cooler as a baseline.

With the active cooler installed on the Pi, running CPU burn at the base CPU frequency of 2.4Ghz, the starting temperature is 41°C and the temperature climbs pretty quickly. Starts to stabilise after about 5 minutes and we have an average stabilised temperature of about 66°C.

With the Pi overclocked to 2.8GHz, the temperature starts at about 54°C and climbs a bit faster. This time it stabilises after about 4 minutes at an average of 74°C. At this temperature, the fan steps up to its highest rpm and this actually brings the pi down to about 71°C quite quickly and it then floats between 71°C and 74°C depending on the fan speed.

I’m actually quite impressed by this result, I didn’t expect this $5 cooler to handle a fully loaded overclocked Pi without thermal throttling.

Testing The Seeed Studio Manufactured Water Cooling Kit

Next, let’s get the Pi fitted to the water cooling kit. Assembly is pretty straightforward, the water block obviously goes onto the top of the Pi with the thermal pad between it and the CPU and surrounding chips.

The aluminium heatsink goes underneath the Pi. This provides a bit of additional cooling but also protects the Pi and prevents any shorts on the contacts on the bottom if you’ve got it on a desk or workbench.

We then connect up the flexible tubing between the water block, pump and radiator. There isn’t a place to mount the Pi to the assembly, so we’ll just leave it on the desk alongside it. This is by design though as the kit can be used with multiple Pi’s running in a cluster as well.

I have to say that this custom water block with a copper base does look pretty good on the Pi, even if it’s a bit unnecessary. It fits well around the taller components and still looks like it will allow you to hook up an SSD to the PCIe port.

Next, let’s fill up the loop with coolant (don’t use water – more on that in my final thoughts) and then boot up the Pi and start our tests.

This block handles the base frequency of 2.4Ghz really well. We start off at about 30°C, then there is a small spike in temperature when the test ramps up but the temperature doesn’t climb much after that. It stabilises after around one and a half minutes at a temperature of about 45°C.

With the Pi overclocked to 2.8GHz, we now get a marginally higher starting temperature of 33°C and again it stays relatively flat, stabilising after a minute and a half at about 51°C.

So this cooling solution works really well and we’re still way under the thermal throttling limit for the Pi. Now let’s see if my DIY version can compete with it.

Testing My DIY Water Cooling Kit

I used thermal paste for my original Pi 4 build, but I want to try to keep the comparison to the manufactured option fair, so I’m using a thermal pad between the CPU and heatsink as well.

I like that my solution has a place to mount the Pi but the kit’s custom cooling block looks a lot better than my simple square block and bracket.

Through testing, similar to the previous solution, this block handles the base frequency really well. We start with a slightly lower temperature of 29°C, then there is a small spike in temperature when the test is started but again, the temperature doesn’t climb much under full load. It stabilises after about one and a half minutes at a temperature of about 42°C. So actually around 2-3°C lower than the kit.

With the Pi overclocked to 2.8GHz, we again get a higher starting temperature of 35°C and it stabilises after a minute and a half at about 49°C, also about 2°C lower than the kit.

DIY Water Cooling Kit Results Overclocked

Testing Results Discussion and Final Thoughts

Both the water cooling kit and my DIY solution work significantly better than the active cooler on an overclocked Pi. That’s not to say that the active cooler isn’t a good option, it is actually quite capable of cooling an overclocked Pi at full load as well.

I’m a little surprised that my DIY solution with the simple aluminium block performed better than the kit’s copper-based block. I assume this is most likely because of a difference in the thermal pads, mine has a really thin and good-quality pad. Another possible reason may be that mine has a greater metallic surface area with the water whereas the kit’s block only has a copper surface on the bottom. The acrylic is a poor conductor of heat.

I think the main benefit of a water cooling setup, besides looking cool, is that they have the capacity to cool a few more than one Pi. Typically a good 120mm radiator can dissipate over 200W. Given that the Pi 5 uses about 12W, even if all of this energy was being converted into heat, we’d still be able to cool over 16 Raspberry Pi 5’s with one pump and radiator set. So while these are overkill for a single Pi, they’re actually quite a good cooling solution for clusters.

There are a couple of things that I like about each water-cooling solution over the other;

The manufactured option has a really good-looking water block and it offers better coverage to the surrounding components.

The cooling block from the manufactured kit is also available individually so it’s possible to buy one pump and radiator set and use it across multiple Pis, which is great!

I do have some concerns that they’ve used copper for the cooling block and supplied an aluminium radiator, so you may run into issues with galvanic corrosion in your loop long term. So, I’d definitely stay away from using water in the loop and I’d use proper coolant to try and assist in limiting this.

My only other comment on it is that the included reservoir’s return line pushes a lot of air back into the loop which makes it noisy. My DIY solution is significantly quieter and you only really hear the fan noise.

I think my stand looks better overall and offers a place to mount the Pi but I am probably a bit biased. I’ve said a water-cooling solution is really better suited for a cluster, so the stand is unnecessary in any case.

Let me know in the comments section below which solution you prefer and also if there is something else you’d like me to test it on.

The DIY LifeTech & Electronics

The DIY LifeTech & Electronics

Where To Get The Pironman 5 Case

Unboxing and Assembly

Assembling The Pironman 5 Case

Pironman 5 Software and Web Dashboard

Testing The Case’s Cooling Performance

Testing The NVMe Drive Speed

Final Thoughts On The Pironman 5 Case

Where To Buy The Beelink GTi 14 Ultra

Tool & Equipment Used

Unboxing The GTi 14 Ultra

GTi 14 Ultra’s Specifications and Interfaces

Taking A Look Inside The GTi 14 Ultra

First Boot and Benchmarking

Gaming On The Stock GTi 14 Ultra

Gaming On The GTi 14 Ultra With An External GPU

Adding The GPU To The GTi 14 Ultra

Gameplay With The GTi 14 Ultra

Testing Power Consumption & Speakers

Final Thoughts On The Beelink GTi 14 Ultra

Parts Used For This Comparison

Pi 5 & N100 PC Setups Being Compared

Testing The Pi 5 & N100 PC

Video Playback at 1080P

Sysbench CPU Benchmark

NVMe Storage Speed Benchmark

GLmark2 GPU Benchmark

Power Consumption

Conclusion & Final Thoughts

Where To Buy The Mixtile Core 3588E

Equipment Used

Taking A Look At The SOM & Carrier

First Boot & Testing

Video Playback At 1080P

Sysbench CPU Benchmark

eMMC Storage Benchmark

AI Object Detection Model

Power Consumption

Final Thoughts On The Mixtile Core 3588E

What You Need To Build Your Own 4-Bay NAS

Equipment Used

Hardware Used To Build The NAS

Designing The Enclosure

Making The NAS Enclosure

Mounting The Components Into The Enclosure

Completed Pi 5 NAS Enclosure

First Boot & Software Setup

Testing The NAS’ Speed

Power Consumption

Final Thoughts On My Raspberry Pi 5 Based NAS

Where To Get The Orange Pi 5 Pro

Equipment Used

Unboxing & First Look At The Orange Pi 5 Pro

Operating System Options For The 5 Pro

Testing The Orange Pi 5 Pro’s Performance

Video Playback At 1080P and 4K

Sysbench CPU Benchmark

NVMe Drive Speed

Power Consumption

Final Thoughts On The Orange Pi 5 Pro

Purchase Links For Components Used In This Test

NVMe Hats Tested:

Test Components & Equipment Used:

Is A Top or Bottom Mount Hat Better?

Performance Testing The NVMe Drive Hats

Final Thoughts On The NVMe Drive Hats

What You Need To Build Your Own FlightAware Tracker

What Is A FlightAware Tracker?

Assembling Your PiAware Tracker

First Boot & Associating It With Your Account

What You Can See Through The FlightAware Web Interface

Final Thoughts On The FlightAware Tracker

What You Need To Assemble Your Case

Assembling The Raspberry Pi 5 NVMe Base Case

Purchase Links For Parts & Equipment Used

Tool & Equipment Used:

Unboxing The Seeed Studio Water Cooling Kit

Outlining The Thermal Testing Process

Thermal Testing The Three Cooling Solutions