UP Squared Is A Very Capable Intel SBC For Makers & IoT

[Guest]

Was it a discount/clearance/something? Because if so, that explains.

It was sold in a discount chain shop, and it's not a name brand machine, and these factors will contribute to the price.

[coder]

Interesting benchmark, but here are several key observations.

Intel Celeron N3350 SoC, 32GB of eMMC storage, and 2GB of RAM.

1. This is their lowest-end CPU: dual-core @ base clocks of 1.1 GHz and Turbo up to 2.4 GHz. They also offer 3 quad-core CPU options.
2. This is using the 2 GB, single-channel DDR4 memory configuration. They also offer a 4 GB, dual-channel memory configuration.
3. This is based on the Apollo Lake architecture, which has been succeeded by Gemini Lake, with its tuned Goldmont Plus core. (and by the way, there's even a newer successor in Lakefield...)

The ODROID-H2 (not to be confused with their ARM A73-based N2) uses the quad-core, Gemini Lake-based J4105, with twice the L2 cache and socketed RAM, supporting dual-channel configs. Also, its clocks are 1.5 GHz (base) and 2.5 GHz (turbo).

https://www.hardkernel.com/shop/odroid-h2/

And with a base price of $111, it would be an even more compelling comparison. It should be noted that the UP^2 included RAM and eMMC, while that H2 price includes neither.

[blueweb]

I want to run this benchmark on an Odroid HC2 running openmediavault, but I am getting an error:

standard_init_linux.go:211: exec user process caused "exec format error"

This is just from trying to run the phoronix/pts docker image, not even the benchmark yet. I checked that other images (hello world, centos) run.

Ideas?

Does phoronix/pts docker image run on 32-bit ARM? Searching that error suggests it could be an arch/platform issue.

[polarathene]

Michael has had an ODROID-N2 in benchmarks before hasn't he? Odd that it wasn't included in the tests, could have easily swapped out some of the existing ones if time / effort for another board was asking too much, I doubt many care for the Tinkerboard for example. But if the article is meant to make the sponsored UP board look like a great buy over everything else, I guess that including more capable ARM boards was a no-no?

Thankfully others have shared their own N2 results here

Interesting benchmark, but here are several key observations.

Those observations are for what exactly? If you went with all the better specs for the board, then you're how much more expensive compared to the mentioned $150? Probably entering mini-PC territory at that point.

For example, I bought one of these around a year ago, price hasn't changed too much from the $190 it's listed for here(I got it from a different store slightly cheaper at $160USD on a special): https://www.gearbest.com/mini-pc/pp_694602.html

Quad Core J3455 (85% better overall perf), 1.5GHz base Apollo Lake also released in 2016Q4 like the N3350
4GB DDR3, 32GB eMMC, Bluetooth 4.0 + Wifi AC. Comes with cooling and a enclosure/case, with a removable HDD/SSD base enclosure(connects via USB-C port internally).

Comes with Windows 10, but that 32GB disk is a bit tight apparently. I gave it to a family member for basic usage, but even though they don't really download anything to it, it appears to have trouble updating due to lack of space(I haven't really bothered to investigate).

Going x86 though, you can have a more reliable experience than with ARM. Plenty of ARM SBCs have numerous quirks and often lack mainline support, even the ODROID-N2(appears to be a common issue with AmLogic SoCs) have issues with USB controller(hardware or closed source driver), most notably causing problems with USB storage and other devices like webcams or tv tuners. Often these concerns aren't thought about when just looking at specs/value, another common USB issue that can be run into(perhaps more SBC specific in general than just ARM SBCs) is underpowered(lack of amps or consistent voltage level) under load or even over time(the power brick can degrade), USB should be outputting 5V, but under these circumstances it can drop a bit too low and cause problems.

It'd be nice to see a SBC that can be powered by USB-PD. Boards like the ODROID-N2 take variable voltage in with DC jack, which is nice since you can get more power from then than say the Pi boards(RPi4 only takes 1.5-3A over USB-C @ 5V iirc, it can't negotiate for higher voltage). With the N2, best you can get from a USB power bank is 2.4A @ 5V through a converter cable to 12V DC plug, not only do you lose amps from boosting the voltage, but it incurs an efficiency loss from the conversion. So you end up with something like ~10W, even when a power bank can output 15W at a higher voltage(requires negotiating for a faster charge protocol like USB-PD or QuickCharge), and that's just with a typical power bank these days.

[coder]

Those observations are for what exactly? If you went with all the better specs for the board, then you're how much more expensive compared to the mentioned $150? Probably entering mini-PC territory at that point.

Again, the base price of the ODROID-H2 is $111. Adding 2x 2 GB SO-DIMMs and a bit of eMMC isn't going to blow much past $150, if at all. So, this reply is for what, exactly?

The aims of my post were to inform folks that:

• The system under test had several mitigating factors that kept it from representing the peak potential of the platform.
• Significantly better versions of this platform (like > 2x as fast) are available, even for potentially less $!

Going x86 though, you can have a more reliable experience than with ARM. Plenty of ARM SBCs have numerous quirks and often lack mainline support, even the ODROID-N2

Is this part of the reply still aimed at me? If so, why? I made it abundantly clear I was talking about the H2.

If you're just talking about the N2 because you also want to shit on it, then I wish you'd make that clear. I'm not going to accuse you of trying to conflate the problems of the N2 with the H2, though the thought did cross my mind.

[Guest]

ARM servers from Amazon were tested right here on phoronix, and they were found to be much weaker than x86 offering. In fact, they had lower value even when taking into account their much cheaper price.

For web services, performance per dollar is not always a goal, as some services have little load, and less performant but cheap solution fits more, when expenses matter,..

[coder]

For web services, performance per dollar is not always a goal, as some services have little load, and less performant but cheap solution fits more, when expenses matter,..

Yeah, I'm not exactly sure what's the point of contention, here. I mean, if you're in the market for a low-cost, compact platform, you're currently limited to Cortex-A53, A57, A72, or A73 cores (plus a smattering of 32-bit ARM cores). So, that's what it'd make sense to benchmark against.

However, if folks want to know the longer-term potential of the ARM ISA,then probably Apple's A12X represents the current best-in-class implementation:

From https://www.anandtech.com/show/13392...icon-secrets/4

What is quite astonishing, is just how close Apple’s A11 and A12 are to current desktop CPUs. ... we see that the A12 outperforms a moderately-clocked Skylake CPU in single-threaded performance. Of course there’s compiler considerations and various frequency concerns to take into account, but still we’re now talking about very small margins until Apple’s mobile SoCs outperform the fastest desktop CPUs in terms of ST performance.

That was concerning the A12. The A12X improved on it, further. You can see how much, here:

https://www.anandtech.com/show/13661...-inch-review/4

The overall SPEC2006 scores for the A12X are 45.95 in SPECint2006 and 58.78 in the C/C++ workloads of SPECfp2006 – representing 2% and 7% increases over the A12.

So, Intel is on notice. Literally, as Apple has already informed them they'll stop sourcing Intel CPUs for Macs (probably starting with laptops) in like 2020 or 2021. I forget exactly when, but it was indirectly confirmed in a recent Intel earnings call with investors.

And although behind, ARM's new Cortex-A77 cores reportedly close the gap with Apple's A12, somewhat. It will be nice to see benchmarks of laptops based on Qualcomm's 8cx chips, which are based on ARM's Cortex-A76.

[Toggleton]

Did a Clean run on the Odroid N2 on the eMMC so that the possible Bottleneck is gone.
Nginx Benchmark did not work. Pearl Interpreter did only run on the A53. The Result Graphs are now comparable.

For Single Core Performance does the Up still win but s922x-b andA311D will run at higher frequencies like the VIM3 by Khadas(A311D - 4x 2.2Ghz A73 andx2 1.8Ghz A53)

https://openbenchmarking.org/result/1908297-HV-1908263AS24&obr_sor=y&obr_rro=y&obr_hgv=Hardkernel+ODROID-N2+4GB

[starshipeleven]

Considering how much a low-end device like a tablet with an Itel Atom costs, this SBC is quite overpriced.

This device has significant more (and better) connectivity, an Altera Max10 FPGA, and it also has a true heatsink so its CPU won't be gated to as low power levels as the one in a tablet.

But yeah, feel free to compare a dune buggy to a truck just because both use internal combustion engines and have 4 wheels.

[polarathene]

Is this part of the reply still aimed at me? If so, why? I made it abundantly clear I was talking about the H2.

It wasn't really specific to you, there isn't a line break on this... oh nvm there is, sorry

Also I didn't realize you were talking about the H2, I thought you were talking about the UP board having configurations for more cores, RAM, etc. Hence I thought that was a weird thing to bring up since at a base of $150, that'd only push the price up further.

I am actually a fan of the N2, I really wanted one, but after doing some research have learned that ARM based SBCs have plenty of drawbacks still so I'm on the fence. So as much as I'd like a decent SBC powered from a USB-PD power bank, it seems for now x86 is worth the extra $50 or so.