Is it possible to build a Raspberry Pi cluster and have it run multiple folding@home projects at the same time?

[wnf]

https://www.youtube.com/watch?v=qfIyuEPIzvU

[muziqaz]

Being extremely picky with the name of the thread: yes. But what I think you meant was, is it possible to run one WU on the cluster. The answer to that is no.
I didn't watch the video, since I'm at work, pretending to be busy.
Raspberry pi is too weak as a standalone and inside of the cluster, since in the cluster you would still need to run client in each individual RPI. They would act as individual systems anyways.
The reason why you wouldn't be able to run a single WU on all the systems:
1) fah likes fast cores, not super many cores. CPU fahcore crashes when folding above 128 threads, it stops scaling past 30 threads + or -. Sweet spot for FAH CPU folding is 16 cores.
2) in home conditions you will not be able to get enterprise or HPC type of network connectivity for that cluster to work as single entity. For the amount you would spend for such set up, might as well get couple of 192 core AMD EPYC systems with several rtx5090s, and buy some beer with the rest of the money.
3) investing in RPI for anything other than network, security, hobby is not the best idea. It is not a compute centric SOC

[PaulTV]

Each step in a F@h job requires all the information from the previous step. That makes that deviding a job across multiple devices isn't efficient - it'd spend way more time with data exchange than with actual computation. Depending on the Raspberry Pi type, it is just fast enough to return a job in time (I believe you need at least an RP4). Running multiple jobs at the same time on the same RP won't work.

Clustering is an interesting concept, it's just that F@h isn't really suited for that.

[wnf]

I was curious, I once saw a model (I couldn't find the video again) of a cluster with 1,000 Raspberry Pi 4s, which would give 4,000 cores, 4000/128 = 31 folding@home projects running at the same time, not to mention the fact that as far as I've seen, the Raspberry Pi allows the use of video cards. Would a cluster of this type be useful?

[muziqaz]

1000 raspberry PIs alone or in cluster would still be 1000 FAHClients on each of the PIs.
GPU on PI is useless, and on top of that it is not supported

[Joe_H]

The interconnects between cluster nodes are too slow for use with the F@h processing. Even with faster ethernet, they are too slow. Processor to processor interconnects on a single chip or motherboard are fast enough. Those are several magnitudes faster.

[wnf]

About GPU in cluster: https://www.jeffgeerling.com/blog/2024/ ... -4k-gaming and https://www.jeffgeerling.com/blog/2023/ ... berry-pi-5

It's a shame that it's not worth using the Raspberry Pi, but it was an idea that crossed my mind. I wanted to build a cluster of 10 units, so it seems that only computers/notebooks can be used.

Just one more curiosity, would these supercomputers like El Capitan (https://tecnoblog.net/noticias/com-amd- ... -do-mundo/) be useful for running multiple projects at the same time or accelerating research?

[Joe_H]

You can use a Raspberry Pi 4 or better for CPU folding. Just not as a cluster. But then again you can't use other systems such as desktops or laptops as a cluster either. There may be no WUs available at times for Raspberry Pi due to size, but they are supported using Linux. As for GPU usage, at this point there has been no work done to create a folding core to run on ARM using ARM based video drivers.

[muziqaz]

About GPU in cluster: https://www.jeffgeerling.com/blog/2024/ ... -4k-gaming and https://www.jeffgeerling.com/blog/2023/ ... berry-pi-5

It's a shame that it's not worth using the Raspberry Pi, but it was an idea that crossed my mind. I wanted to build a cluster of 10 units, so it seems that only computers/notebooks can be used.

Just one more curiosity, would these supercomputers like El Capitan (https://tecnoblog.net/noticias/com-amd- ... -do-mundo/) be useful for running multiple projects at the same time or accelerating research?

If one has a budget for El Capitan like system, one will adapt their workload to that system.
FAH has no budget for El Capitan type of system and it is not built for this type of system, thus El Capitan and other supercomputers are useless to FAH