I tested how the 5B's RK3588 fares against an Intel N100!

Would of been interested in GPU tests as CPU perf landed where I sort of expected, but haven’t a clue of how the N100 GPU performs.
Great review though and very interesting.

I think Rock 5B (not all RK3588) is more expensive than N100 from an “overall package” prospective. I am looking into another soft router, and I think N100 / Gold 8505 are better choice instead of cheaper alternative based on RK3568 even. I would consider 8505 because a 4-port i226 N100 and a 4-port i226 8505 cost around the same now (around US$110 barebone), but N100’s DDR5 is still more expensive than 8505’s DDR4 per GB (around 80% in where I live for 16GB), let alone N100 normally take one SO-DIMM only (up to 48GB now) while 8505 takes 2 (up to 2x32GB). But of course, 8505 power usage is quite a bit higher… I saw something like N100 at around 30W full load, and 8505 at around 50W full load.

Where you from @enoch as on a 1st google
https://www.amazon.co.uk/Crucial-CT2K32G48C40S5-2x32GB-4800MHz-Laptop/dp/B09S2QLBWC
Have a shop around as just of late I have and noticed its getting to be about parity I guess like ddr3 did when it got scarce DDR4 will become the more expensive

Retail in my area is kinda crazy, I dunno why…

Thanks for reminder, I can get directly from Amazon… take Crucial as an example, DDR4 16GB is US$32, and DDR5 is US$41, really not “as bad”, that DDR5 is just about +28% more

1 Like

I dont think this is a fair comparison but even if it is. Its more or less 50/50 but then again can u find credit card sized boards running intel n100? And knowing the Radxa Pi X fate of loosing stock to survive and mantain sales happening and knowing the US politics over China chips would that chip last for adoption ? I think these are the main questions here / issues to have over the dick chip measure debate.

Rockchip made us an great upgrade with the RK3588 our tiny boards are running arm64 fast.

I still think that RK3588 beats these goals.

Not fair in what way exactly? Just because the N100 device is bigger? I’m hoping you actually read what I wrote and you’re not just skimming it :grin:

2 Likes

I can add some GPU bits to a todo list to come back to, though I didn’t initially include anything as it would have been a little more difficult to do an actual comparison between the 2 (as all of the hoops to get the GPU working nicely enough on the RK3588 would just be proving my point in other areas… :smile:)

Was there anything in particular you were interested in seeing from the N100’s GPU?

Its just my opinion over the need for a comparison.
If the initial idea was more to make sure and show proof that intel n100 is better you nailed it.

But… imo its just not comparable and as I mentioned the board size capability also matters therefore the arm64 arquitecture.

We cant ignore those pros. Anyway yes its just a valid comparison to take out some ideas but from my experience the RK3588 doesnt disappoint.

Great comparsion. There’s no denying that these x86 boxes are far superior to ARM SBCs in terms of out-of-the-box usability.

But my concerns about these low-power x86 Intel chips are:

  • Intel applied various software-based workarounds to mitigate security vulnerabilities at the expense of performance on 6-11th gen chips. Have to guess that the same thing might happen on these 12th gen chips in the future. I was bitten by this on my last Intel laptop.

  • These random boxes from AliExpress are having trouble getting long-term UEFI BIOS and vBIOS updates from Intel and manufacturers. Basically they are one-time sales. This is one of the reasons why competing products from the likes of Intel NUC or ASUS are more expensive.

  • The ability to run fanless is a good selling point for me. These x86 boxes usually use custom cooling modules, and once the fan breaks out of warranty, you will have a hard time dealing with it.

3 Likes

So my post wasn’t just to show “Hey look N100 is better, ARM is dead, long live the King”, far from it. Your point about size feels a little laboured given that that’s not the main concern for every user and I’m not sure what you mean by the arm64 architecture when it comes to the size.

I really like the ARM platforms, you can see that from my website, it revolves around ARM SBCs and this is the first mention of an x86 platform on there. This piece came from the fact that after launch of these higher-end RK3588 boards, there was considerable disappointment on the software side of things and people spoke about selling them to buy mini x86 PCs to replace them (you can find many such posts on this forum) and during the Raspberry Pi shortages, a lot of people were redirecting buyers to SFF PCs as alternatives due to pricing/availability.

With the Pi 5 launching and there being frustration again over the availability/pre-order situation, I thought it would be a good time to finish off a piece like this, though using the RK3588 as the comparison due to it being the best SoC available in the area.

I make a specific point of not saying one or the other is the best option in my conclusion as there is no “best” for everybody. You may be happy to spend $150-250 on an RK3588 board and the necessary equipment before spending hours making things work and navigating documentation but someone else might not. In their case, alternatives like this are an extremely compelling purchase.

But hey, I’m not trying to cause any arguments, what’s right for you isn’t going to be right for everyone else (and vice versa), I just didn’t understand the “not fair” part of your initial reply and wanted to know why you felt that as the N100 not being available in a credit card sized format didn’t feel like the strongest argument for it :smile:

2 Likes

This piece came from the fact that after launch of these higher-end RK3588 boards, there was considerable disappointment on the software side of things and people spoke about selling them to buy mini x86 PCs to replace them

That’s why ARM, Linaro, Rockchip and Radxa started work together to push the ARM SystemReady for RK3588, ROCK 5B as the target platform.

3 Likes

Yup, and over the last year there have been some good progress on the software side but it’s still going to take more time.

Now this little back and forth has happened, hopefully you don’t mind that I posted this here, I’m by no means trying to talk down on Radxa with this piece. I have 5-6 Radxa boards sat here and the original Zero was one of the first “alt” SBCs that I purchased, you’re just caught in the crossfire a little as the 5B was the only RK3588 board I had full benchmark data for at the time of composing this :smile:

I think the IGPs via intel have improved as they had that period where they looked like they might start doing decrete cards that to be honest dunno if or not.
But they did start to beef up there GPUs and glmark2 will run on both and likely you can think of others such linux based gpu ml tests.

I have zero idea of where they both land, not a clue and it would be interesting.

Intel made several discrete cards, they are called the “Arc” line. They’re not great, the Arc A770 still often falls short of an RTX 3060ti, whilst costing around the same. It also falls short of the RX 6650XT whilst costing more.

I don’t quite understand your pybench results, they’re listed in milliseconds but even the i7-8565U took 50 seconds according to pybench.

Running pybench on my own 5B gave a benchmark time of 1m 51.07s, are your results for 1/3 of the test?

I normally consider (trade off) in the order of:

  1. target usage / application
  2. hardware configuration, that’s derived from target usage
  3. size
  4. power = heat generated = noise from fan
  5. availability of accessories

say for now I need a soft router for a secondary location, and Rock 5B would not be a good choice here, though a N100 / 8505 mini-PC will do (e.g. you can search for G31F 8505 which I am considering a variation of it); however, I still need a “desktop like” system that could let me use a more recent and “vanilla” Android “occasionally”, that means Rock 5B would be a good it here running openfyde…

then we also need to look into the cost side… a “full featured” RK3588 mini-PC should not be cheaper than a N100 or 8505, just smaller and consume less power… and then we also need to look into the software (driver) side of things, like if I am “fine” with out of box experience? or I want customization? in this case, a x86 platform is much easier normally, and again I still have issue with Rock 5B not starting my Pi compatible 5" DSI touchscreen, so for a typical end user I think this will be something to avoid in the first place, so it really depends on the trade off I am considering, which may include buying multiple devices instead of just one, even though “device consolidation” would be something really nice

2 Likes

Or there’s a much more straightforward answer and I’ve made a typo :stuck_out_tongue: Let me double check!

Yeah exactly as the Iris and IGPU ranges got some of that work backported and they have been beefed up.
Arc wasn’t great but I have no idiea apart from synthetic such as supposed GFlops/Pixels.
What a Intel IGPU vs Mali-G610 with actuals from maybe GLmark and some GPU based ML that just theoreticals would be interesting.
The more tests the better as have never really seen it done, no-one ever seems to bench Intel Igpus

2 Likes

great comparison i’m still looking for a way to run proxmox on rock 5b with router/firewall/NAS

for cpu inference on AI stuff i expect the N100 to be faster(on my tests the ARM board can’t make use of even 50% the ram bandwidth

i expect the N100 to be faster(on my tests the ARM board can’t make use of even 50% the ram bandwidth

There’s an issue that was noticed with the DDR v1.16 init code that makes RAM super slow (at least DDR5), but we don’t know if the source of the problem was well identified yet. In short, there was a setup (1.11 or so) with where DDR5 was set to 5472 MT/s that was apparently fine, but for “stability reasons” (as indicated in the commit message) it was reduced to 4800. But while doing so, something else was clearly changed because the RAM’s performance was cut by 2.5 or so (for DDR5 as well). The problem is that it’s the same file that brings DDR4 and DDR5 timings and since it’s unclear what was changed I don’t know if it’s possible that some rock5b are shipped with this version and are affected with the issue. At least rock5-itx’s ddr5 is seriously affected, and my old rock5b is running perfectly fine (I don’t know on which version). Maybe the issue only affects ddr5 (we can hope so), maybe it affects something else that is used to build these binary files and could affect the two versions.

I guess running sbc-bench on your machine and comparing the results with other public ones will tell you if your machine works as expected or is slower.