If I stripped my old laptop down to a single PCB, it certainly wouldn't boot, it would be missing RAM, and storage. My PCs would have no RAM, no CPU, no storage, potentially no GPU, and most of the servers would be the same. The IPMI/iDRAC etc would boot up I guess if we count that, and we'd potentially get into the BIOS if a device has soldered RAM, but M.2/SATA attached storage or something?
We could debate about strict definitions all day, but I think the vast majority of people differentiate between a Raspberry Pi 5 (not a great example given it needs storage, pick anything with eMMC if you truly want everything on a single board) running a full Debian-based OS, and a Pico running MicroPython (or whatever your poison of choice is) and one task at a time, at least I do when it comes to this kind of thing!
Hmm, I'm not sure if I'm missing something but that 1st comment is what I'm doing. I have 3 different sized Deepseek R1 models (1.5, 8, 16) and they run on each board that can handle them and then the data is reported.
For the 2nd, the file I grabbed initially was https://github.com/geerlingguy/ai-benchmarks/blob/main/obenc... - which I now notice wasn't modified in his repository, so I can check that out, but either way, the same version has been tested across everything thus far.
The balancing of wanting to draw attention to sbc.compare and allow the post to stand on its own feet has failed it seems :D No worries, the feedback is good and I'll see if I get time to go back and add a summary for this, otherwise, I'll definitely take it on board for the next one(s).
Thanks! See my reply below to the other user on this, I think the links to sbc.compare in the article have a lot of the data that you're looking for. I didn't really want to duplicate all of the data I have on the other site, I was more giving a quick recap, with links to the mass of data if people wanted to go that far. Some you mentioned like the dimensions are in the database, but I've not yet exposed them to the frontend.. Ever growing to-do list of life!
For that I'd really point you towards the links in the article, as that was really what it was supposed to do :D As I mentioned to another person in the comments, this was literally supposed to be a recap of the boards I tested, with links to in-depth benchmark results for each of the boards, with the ability to then filter and compare against around 100 other SBCs that I own and have tested in a controlled manner. Software support is a very tricky one though, sadly. I have a few ideas for little things like being able to search for Armbian support, but others would need a long hard think!
Hey! Author of the post someone linked here. Fair comment, though this wasn't really meant to be a review, or "go buy this!" type of post, it was more to highlight what I tested from the boards released in 2025 and share the results to those benchmarks via sbc.compare
Armbian do a great job of handling support for a whole host of boards (including most I included in this list), so you'll usually have Debian/Ubuntu-based flavours. Vendor kernels and vendor supplied images will be hit and miss. Mainline Linux support is a flag you filter by on the benchmark comparison site linked in the article, but it's a difficult one to keep up to date and define exactly. It could have some kind of support, but miss out on display functionality, or WiFi yada yada. What would we then class as having mainline support? All hardware etc functioning? If so, very, very few will meet that definition.
I get the desire for the information, and perhaps I should have envisioned these types of questions, but all I initially meant for the post to be was a recap for people following me to see which boards I'd tested that were released last year :D
Under the Test Environment collapsible I have the OS version and kernels tested at least, but I could definitely look at adding something extra to help make things more obvious on that front
About 65 boards in I realised there was a slight error with how the idle power consumption was being recorded, so I had to scrap all of that data :( The last 15 or so do have this, but I made the decision to backfill that data as and when I need to check something on those boards, or just on the next round (I plan to update every X months or so, assuming there are worthwhile updates)
Realistically, I don't think any of the tests I have would run on something like that, I'd need to look at different types of tests, and then they wouldn't really be comparable with the rest. Though, I don't think most people would really be thinking "Hmm should I get a Luckfox Pico Mini A, or a Raspberry Pi" but the feedback surrounding these smaller boards has been interesting, so never say never, but I don't think it's within the scope I had in mind just yet.
I'm discussing some other bits with Banana Pi so maybe I can ask about that, I'd need to look into the device to see how viable it is, let's see! I can definitely add the "How did you get this board" disclaimer there somewhere
I don't know if I'd release everything as is, there's a lot of hard coded nonsense that I'd need to filter out and after testing 80 boards, I have a bunch of improvements I want to make. Opening up the test scripts to others isn't/wasn't a high priority item, as I didn't even plan on making this site initially :D
Room temperature data is also nice, but I'll need to draw a line somewhere, on average these tests take around 3-4 hours to run, and with 80 boards right now, best case scenario is 10 straight days of testing (in reality, more with setup and.. life). I could use data on the most popular boards and only do a handful, but I probably have to limit the scope at this point and clearly state the testing conditions and how it may differ in your real world setup.
I can definitely do more to expand on the testing process though, and I can look at sharing the exact commands that are run with my automation so that users can at least mimic the benchmark runs if not the full automation. I really appreciate the responses and feedback, though, so thank you!
It's a little bigger than what I imagine you're looking for, but the Radxa 5 ITX (not the ITX+) has 4 SATA ports, PoE+ (via a separate module), 2x2.5GbE, and the RK3588 SoC - I've been testing OpenMediaVault on it for the last couple of weeks and it's been pretty solid. If you're fine with M.2, the Radxa ROCK 5B+ has 2 M.2 M-Key connectors, the same RK3588 SoC, and PoE via a HAT bought separately.
Otherwise, there are dual NVMe HATs for the Raspberry Pi 5 but you'll be splitting a single PCIe Gen 2 lane across both drives (unless you go for the much more expensive HATs that have a Gen 3 PCIe switch on them to then share a single Gen 3 lane across them!)
If you could share a little more on what you actually mean, that would help a ton. I’m not a developer, I put this together as I went, learning along the way. It’s not perfect, and I’m aware of some issues, but if you’d be so kind to expand on “pretty terrible” then maybe I can see if it’s already known, or something I should add to the list.
Hmm, the search by SoC should work already, but there have been a few regressions with the search functionality that I need to fix it seems. Notes on the other bits too, it’ll take a couple of weeks I imagine but we’ll get there!
Raspberry Pi products can be found in a lot of industrial/professional setups, especially things like their compute module range(s) that can be dropped into off-the-shelf, or custom carrier boards. Others can be a bit more hit and miss, though Libre Computer are up there with great software support, and will have their own compute modules available soon.
I hear you! Like I mention to others in the thread, this functionality is already there, I just need to finish populating all of the data. To get over a mental block I had to decide whether I'd continue trying to add each feature before an initial launch (and be there forever) or give myself a shot of motivation by getting an initial performance comparison feature set out there and iterate as I go along.
As soon as I have all of that data in there (I think I'm at around 30-40% so far, the initial batch of testing has been a slog, data entry for this took a back seat) I'll be enabling that option and it will all be there to view on comparison pages, and search for to help find/compare on a deeper level.
I do have a ko-fi available in the footer of the website, and if someone wants their donation to go towards a specific piece of hardware then I'll most certainly do what I can to honour that, or return it if I can't.
On the samples from vendors front, yeah, that's definitely a concern, though in the years I've been working with manufacturers and had samples from them, I'm fairly confident that no "golden sample" binning is going on as I've had quite a few shockers, hah! Not to say that it couldn't/wouldn't happen, I wouldn't want to introduce that doubt. Perhaps, like my review website, I should add a quick note/indicator of whether a board was obtained directly from a vendor, or if it was purchased.
We could debate about strict definitions all day, but I think the vast majority of people differentiate between a Raspberry Pi 5 (not a great example given it needs storage, pick anything with eMMC if you truly want everything on a single board) running a full Debian-based OS, and a Pico running MicroPython (or whatever your poison of choice is) and one task at a time, at least I do when it comes to this kind of thing!