Odroid N2

[guldhammer]

Hi.

I was considering to build a cluster with some Odroid N2 (4gb coreelec edition)
Could this work out?

Regards
Michael

[toTOW]

Folding@Home doesn't support the ARM architecture. Only x86-64 is supported on Windows/Linux/OSX.

[guldhammer]

Great!!

[liminal]

How about comparing Odroid benchmarks for an exploratory sense of how viable the idea might be.
https://browser.geekbench.com/v5/cpu/search?q=odroid-n2

Do let us all know how your follow through goes. Whether Odroid clusters could work well for folding is out of my depth, however, what about this alt download?
https://download.foldingathome.org/rele ... _arm64.deb

It's a little surprising how folding has kept lively support for RPi and the like, even though the PPD numbers per watt or per retail cost of the kit don't seem all that impressive compared to mass market number crunching hardware. Someone correct me here? I seem to recall encountering a forum post considering Odroid as high ranking contenders among SBCs.

[JimboPalmer]

Here is a web site for the board, it states Ubuntu 18.04 64 bit is the supported OS (which should support the .deb file format)
https://www.hardkernel.com/shop/odroid- ... gbyte-ram/

F@H support seems to match a Power * Supply concept; your supercomputer may be useful but if there is only one of them, it is not worth writing a client for. RPI has more volume, than Odroid, that may mean more work done. I am not seeing any reason the same client won't work for both.

I do not think F@H is writing for the RPI specifically, they mean to be an ARM port.

https://foldingathome.org/2020/11/24/ne ... m-support/

[MeeLee]

Odroid makes x86 boards, as well as ARM.
For ARM you just need to get an A70 series compatible CPU. I believe the A50 series CPU still won't work.
It'll be dog slow though, like ~1-3k PPD or so...
Not sure if folding will work with a BIG-LITTLE architecture.

[PantherX]

IIRC, you can fold on the BIG.little architecture but the overall speed will be limited by the speed of little cores due to the way the threads sync.

[MeeLee]

IIRC, you can fold on the BIG.little architecture but the overall speed will be limited by the speed of little cores due to the way the threads sync.

Was this tested on Linux or Android?
Android usually shifts the tasks to the slower cores, which could crash.
Is Linux using all 8 threads or only 4?

[Joe_H]

There is no folding using Android, just Linux. The testing done was done with the Linux ARM client, only the A8 folding core has been created to run on ARM. What I recall from that is that the low power cores can contribute, but some tests running on just the "Big" cores did run faster than with all cores active.

[bruce]

FAH assigns a variable number of threads. The processing speed is generally limited by the workload assigned to the slowest thread. If there are a mixture of fast and slow threads you have to consider whether it's faster to run a small number of fast threads or a larger number of mixed performance threads. A few fast threads will finish first and then have to wait for the slow threads to catch up.

[MeeLee]

From my data on Android OpenCL tasks, it does appear that when the small cores are running, the bigger cores lower in performance.
Possibly a cooling issue.

The benefit of ARM is that it has much faster RAM access than x86, so it doesn't need as much L-cache, which helps keep the CPU cool.
In most scenarios, 14nm or smaller A50 cores can run without any cooling (at 2Ghz).
But A70 cores do need some sort of heat sink to stay cool.
Even the more modern Neon cores running at 10nm. Instead of lowering TDP, manufacturers raised the speed to 3Ghz on these CPUs, and they run just as hot.