Nvidia Jetson nano super

[thiblec71]

Can folding at home run on the new Nvidia Jetson nano super?

[Joe_H]

You could probably run F@h using the ARM version on Linux and do CPU folding. However they have not created an ARM version of the GPU folding cores, so no GPU folding. I don't know how the processing power of the ARM CPU in a Jetson Nano stacks up compared to other ARM processors. It would need to be at least as fast as a Raspberry Pi 4 to be useful.

[GOTspectrum]

You could probably run F@h using the ARM version on Linux and do CPU folding. However they have not created an ARM version of the GPU folding cores, so no GPU folding. I don't know how the processing power of the ARM CPU in a Jetson Nano stacks up compared to other ARM processors. It would need to be at least as fast as a Raspberry Pi 4 to be useful.

Is that on the cards at any point?

We are seeing more and more SBCs with either dedicated GPUs or PCIe connectivity, meaning you can add GPUs. I would certainly love to be able to run a GPU on arm, seeing that CPU and GPU folding at the same time reduces overall PPD, it would mean I don't have to buy expensive x86 HW as SBCs continue to advance

[Joe_H]

Is that on the cards at any point?

I haven't heard anything about them being looked at so far. Part would depend on whether they become common in usage. The other part is whether the GPU section would be useful for F@h. My brief look into that is it looks like a cut down Ampere chip from that used in the RTX 3050, but optimized for tensor operations such as used in AI. It still has the vector section which is used by F@h CUDA processing. Against that is the additional programming and testing time to create and support an ARM based GPU folding core. But if these became common enough and powerful enough, then the effort might be considered worthwhile.

[GOTspectrum]

Is that on the cards at any point?

I haven't heard anything about them being looked at so far. Part would depend on whether they become common in usage. The other part is whether the GPU section would be useful for F@h. My brief look into that is it looks like a cut down Ampere chip from that used in the RTX 3050, but optimized for tensor operations such as used in AI. It still has the vector section which is used by F@h CUDA processing. Against that is the additional programming and testing time to create and support an ARM based GPU folding core. But if these became common enough and powerful enough, then the effort might be considered worthwhile.

That's fair, although I was talking about arm-powered devices in general, not just the jetson. We are seeing more and more SBCs with PCIe, such as m.2 slots, and more arm-powered systems in general. Not that I think we will see them "take over" any time soon, but, they are becoming much more common.

Would be a nice "side-project" for someone to tinker with, just in case ARM becomes more commonplace. But I don't know how core software development works, so... Maybe not a possibility