Folding Forum

Does fah projects support ARM CPUs? Id like to use my Jetson Nano.

ARM CPUs are not supported. Sorry.
Although the Jetson has a good GPU, there are probably too few of them around and available for it to be worthwhile to develop the CPU side client code.

If you're really ambitious, Sony did open source their Android FAH client! You can learn more on the Android client from Sony forum page: viewforum.php?f=100

As I know Android folding project was closed in 2018.

Even then, the Jetson is basically a quadcore ARM CPU running at (or around) 2Ghz (equals a dual core 3Ghz CPU of ~1500PPD, and barely makes the deadline), and a GT730-like GPU (384 cores at a low frequency).

It's like both CPU and GPU wise, it would be the absolute minimum to make the deadlines.
Not recommended for Folding.
We're all waiting for Nvidia to up the core count by 20-100x, and allow access to 2x PCIE x16 slots.
Then it might be worth it. (for like $500).
But the $399 for a board that's essentially a $35 TV box, with a $35 GT 730 pasted to it... I wouldn't recommend buying this.

Besides the hardware, You will see CPU work began when Intel Partnered with F@H

ATI (now AMD) partnered to get GPUs working and Nvidia partnered to get their GPUs working. Sony made PS3 and Android attempts

https://foldingathome.org/about/partners/

I do not think ARM has the deep pockets to fund development.

ARM is nothing but a license company, they don't actually manufacture anything.
They hope companies like AMD would buy their licenses, and manufacture their products which basically means free money on a patent.
For a while AMD did this with their Opteron A-series CPUs, but they didn't last long.
Limited to 4 or 8 cores...
Which licensing from ARM is a dumb thing to do, as ARM processors are basically derived from RISC processors, which are open source and free for all to use.
RISC V is not that far off ARM, but documentation and application of it, isn't very widespread, and there still are some differences between the two.
What AMD needs, is get the license for a server chip that would bust the market (like a 64-256core/128-512 thread CPU), that they know would sell like hotcakes.
Trying to get a license to modify ARM processors to run 8 or 16 cores, isn't going to cut it anymore in this market.

ARM processors may very well die soon, if they not somehow find a way to increase core count to remain competitive in the market.

That's why I believe AMD basically crushed any hope on ARM becoming a player in the server market, with their 7nm Epyc, Ryzen, and Threadripper CPUs.
Using about the same power at 7nm per core as ARM (at 12nm), but they run about 2,5 to 3x faster core speed (not to mention the increased performance per Mhz of x86 vs ARM).

AMD had no choice but to inject some spectacular product in the server market, to keep ARM out of that market.

WCG has ARM/Android project for current pandemic. Have a 5000￥ cheap android tablet crunching it on three cores ... one WU around 30h. But works ...

OK, so AMD managed to crush the ARM server market. (not 100%, but close)

I suspect the same logic applies to FAH crushing the DC market where speed is a factor. Yes, a cheap android tablet or phone can accomplish something in 30h, but you have to ask yourself whether adding that capabiiity to FAH would make a significant difference in the long run. FAH doesn't thrive on lots of independent WUs running on hardware that produce N GFLOPS compared to fewer independent WUs running on hardware that produces M GFLOPS where N<<M.

Just how many TFLOPS or GFLOPS are being used productively on your tablet? What FAH hardware is comparable and what Timeout/Deadline do you think your tablet would be able to accept?

Bruce, administratively, it is worse than you imagine. The flavor of SIMD (think SSE2 or AVX_256) for ARM is called NEON, but Double Precision 64 bit floating point math is optional in NEON.

So every single user of ARM will be a lottery if they can fold on F@H. They won't know in advance if they can fold, nor will we be able to offer any help but 'Sorry"

https://en.wikipedia.org/wiki/ARM_archi ... SIMD_(Neon)

Unless of course an Armcore was written that didn't use FP64 … Yes is would be a problem as things are architected/"procedured" today - and I am in no way saying it should happened - but something like that is just a "simple" challenge for the devs

Neil-B

How do you FORCE the biochemists to choose less accurate science over more accurate science? The trend in F@H has always been toward more accuracy as they can afford it.

Who said anything about FORCING anyone to do anything ?????

Depends on what type of folding core they target the device with. The Gromacs based CPU folding cores have only been using FP32 calculations the last decade plus, though they did do some experimenting with higher precision earlier on. It is the OpenMM based GPU folding projects that have required mixed precision lately, and thus need FP64. But under under other circumstances a project for GPU folding could be set up and just require FP32.

So if, and that is the big question, they decided to go for folding on ARM, they can either choose to target for the lowest common specified functionality, in NEON's case FP32, or they would have to get some device ID to allow specific assignments based on presence or absence of FP64.

I'm not saying the FAH should jump on the ARM track; more the opposite.
If I have an ARM device I put it on a different (and in my eyes also worthy) project. And keep my Nvidia stuff etc on FAH; since WCG is not supporting GPU these days anyway.

So for me its simple: I choose both concepts and assign matching hardware to each.