NVidia Jetson Nano (unsupported)

[wcbutler]

Anyone gotten one to work? Didn't spend a ton of time but got frustrated with the installer even though I feel like I've installed the DEB file in Ubuntu before.

[JimboPalmer]

In theory, it should not fold. All current folding clients use the x86 instruction set, often x86_64.

The Jetson Nano uses an ARM instruction set, for the A57 CPU.

While the GPU is a Maxwell design that would work if ARM software was developed, again, all the GPU code assumes a x86 host CPU. The GPU may be so weak it cannot complete a WU in the allotted time.

Sony wrote an Android client for ARM, (not linux) but it never supported GPUs and is currently inactive.

[strombergFs]

can somebody explain or show the difference between a NVIDIA grafic card which is good for folding here and the Jetson nano, where the peoble say it would be too slow.... I get it that ARM CPU is not supported..but why would be the Jetson nano (the MAXWELL GPU with the cuda cores) too slow?

[Rel25917]

128 cuda cores with no mention of core speed doing .5 teraflops(fp16) vs 2560 cuda cores on a 2070 super with 1600mhz base clock doing 9 teraflops(fp32) or 18 teraflops (fp16). It looks like an old gtx 560 can do 1 teraflop(fp32). So about 1/4 the speed of a 560...

[Joe_H]

And the ultimate lack on the Jetson Nano, apparently no support for FP64. The GPU folding projects currently require FP32 for most calculations, and use some FP64 calculations where needed to maintain enough precision.

[ori333]

Hello, I have a Jetson TX1 and would like to try. Is there a FAH version I can build and try to get it working?

[Joe_H]

Hello, I have a Jetson TX1 and would like to try. Is there a FAH version I can build and try to get it working?

At this time, no there is no version that can be used for the ARM based Jetson. At most, running Android there are some monitoring apps that might run on one and that could link to other x86 based systems to monitor what they were doing on their F@h clients.

[MeeLee]

The nano is pretty much a joke at the price you pay for it.
You won't make the deadline with it, both on CPU or GPU if it was ever supported.
It's great for automotive entertainment systems, but not for deep learning, or folding.
Is like cleaning your home with a toothbrush.
I was seriously disappointed with the performance of it.
What they instead should have done, is modify the board to work on 12V, and have a PCIE x4 port that could click on the back of a GPU.

Performance wise, it's not faster than any Intel Atom or Celeron powered laptop.

[szopen]

There is alternate folding project, working on a Jetson Nano:
https://foldforcovid.io/
it is BOINC based, but does not use GPU; only CPU, so is has questionable performance.