FAH for Raspbian/Raspberry?

[P.U.Baer]

Hello,
i own several, temporarily unused Raspberry Pi 3 and Pi 3+ with Raspbian OS.

I would love to let them contribute to Folding at Home, but i am not too experienced in computing.
Is there any experience oder information, how to run FAH on Raspberry Computersystems?

Thanks for help and your infos

P.U. Baer

[davidcoton]

Sorry, there is no client for the Raspberry Pi.
In any case Pi computers are very low power compared with average desktops, and would take too long to process a WU.

[P.U.Baer]

Even the raspberry Pi 4? Well, my desktop seems to be an old one, too...

[JimboPalmer]

[I am just a volunteer like you, these are just my opinions]

Sony paid for and wrote a project for Android on ARM.
After they wrote it, ARM started using multi CPU system using big.LITTLE, which confused the Sony code.
Even new Sony phones did not work with F@H.
Sony put the code into the public domain and abandoned the project.
The code is still out there, doing little.

There are huge issues to getting this code in your Pi.

Android vs Linux.

Single threaded CPU vs SMP big.LITTLE code

CPU code vs GPU code.

https://www.androidpolice.com/2017/10/1 ... en-source/
https://en.wikipedia.org/wiki/ARM_big.LITTLE

(One issue with Pi, is that all previous F@H projects have had a corporate sponsor, Pi has no 'Deep Pockets' funding source)
https://foldingathome.org/about/partners/

[owen]

Still, I think this is seriously a missed opportunity for FAH.

I've got two computers running FAH (One Linux Server and one Mac Desktop).

If there were an FAH port for PI, I've got 20 I could throw at the task tomorrow.

I realize that the PI 3 and earlier are relatively slow compared to desktops, but they're not that much slower than a 5 year old pentium-based system which FAH still supports.

I realize some people are hardcore into this for the points. Many of us aren't.

Now, with the advent of the PI 4, I think you're looking at something which is not nearly as underpowered as previous PIs and Arms are only going to get more powerful.

ARM Linux will continue to be deployed on ever more powerful machines. Also with the high probability that Apple will be making an Intel->Arm switch, that's another good reason to look at an ARM port.

[AlphaSerpentis]

I myself have a Raspberry Pi "Supercluster" consisting of a Raspberry Pi 3B+, and two Pi Zero W's.

I'm fully aware that they're not the most powerful systems, but if we can gather people's unused computational power from all of the RPIs around the world, it could make more than just a scratch on the surface.

As previously mentioned, some of us are here for the points, but some of us are also willing to put as much computational power as we can to better help Folding@Home's work.

[davidcoton]

It's a balancing act for F@H. When is it worth the investment in a recompiled core, client, and control, all on a new architecture, to capture a group of nodes that ate low-power but numerous? How big is the potential Raspberry Pi userbase? Total Pi sales are known (roughly), and the compute power is known, and the figures do not stack up well. Some work is better suited to CPUs than GPUs, but CPUs are way behind GPUs in the FLOPs available for folding. And the Pi has no useable GPU for this purpose. So it is right at the bottom of what F@H supports, the Pentiums are only supported because the software is available and doesn't need to be written. There is the added problem that new F@H software is likely to be 64-bit, while the mainstream Pi is still 32-bit (I know that will change, but not today).
So the cost of catching this (not so good) opportunity would be diverting the F@H programmers from improvements to the mainstream software -- which would lose far more in processing capability than the Pi would bring in.

Sorry to sound negative, I own and use Pis in the right place. Just not for folding.

[JimboPalmer]

Still, I think this is seriously a missed opportunity for FAH.

I realize that the PI 3 and earlier are relatively slow compared to desktops, but they're not that much slower than a 5 year old pentium-based system which FAH still supports.

ARM Linux will continue to be deployed on ever more powerful machines. Also with the high probability that Apple will be making an Intel->Arm switch, that's another good reason to look at an ARM port.

Correct, if you could convince Apple to fund an ARM core, there is a funding source to rival Sony. "a 5 year old pentium-based system which FAH still supports" uses code paid for by Intel, it cost very little to support it once written.

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

The Sony code is open source, nothing prevents you from re-writing it for the Pi. What you will live on while you write such an application is an issue, as is programming skill, but you are discounting those issues for others, so why not you? Me, I am hoping for deep pockets and corporate skill levels, no part of F@H has been completed without them in the past.

[owen]

Given that there's already a Linux client, a first approximation should be as simple as a recompile using an ARM compiler. If the FAHClient source code were FLOSS, I'd take it on.

Unfortunately, because there's apparently some issue with people cheating on points, it's not. (Is there some value to these points of which I am unaware that makes it worth cheating the system to get them?)

Since I can't get the source, I can't recompile it myself. I don't have the deep pockets to fund someone else doing development, but I'm betting that there are plenty of talented developers that would be willing to do an R.PI port if they could.

[JimboPalmer]

Given that there's already a Linux client, a first approximation should be as simple as a recompile using an ARM compiler. If the FAHClient source code were FLOSS, I'd take it on.

Unfortunately, because there's apparently some issue with people cheating on points, it's not. (Is there some value to these points of which I am unaware that makes it worth cheating the system to get them?)

Since I can't get the source, I can't recompile it myself. I don't have the deep pockets to fund someone else doing development, but I'm betting that there are plenty of talented developers that would be willing to do an R.PI port if they could.

I do not think it is the user interface that is a hang up, I suspect it is writing a Core for NEON. (It is possible NEON is being replaced by Helium)
https://developer.arm.com/architectures ... or-armv8-a
One assumes that Sony got some part of this done, but I am not an ARM programmer to know, I do know their user interface did not resemble F@H. (I assume their version did not issue points because points are integers and their results were fractional)

I do not pretend to understand why some people need to cheat the system, I really don't. I do know that some feel the need to be on top without doing any work. (Even here, where the PC does all the work)

If my theory is right, and it is the Cores that are difficult and the UI is just a recomplie, you could work on the open source science code and develop the science in ARM.
http://www.gromacs.org/ (some hints that recent ARM is supported via gcc)
http://openmm.org/
https://simtk.org/plugins/phpBB/viewtop ... rt=0&view= (Someone trying to get OpenMM on a Pi)

[davidcoton]

The problem is bigger than just recompiling to a Pi-friendly ARM code. Pis have used FOUR different processors in eight years. Anything written for Pi may have a limited life. Every re-write of the science code needs to be validated by the researchers. So even given the availability of a RPi core and client, I doubt the potential compute power would justify the researchers' time.

Would you target CPU or GPU (the former is not very powerful, the latter may not even be possible -- different architecture, no CUDA or OpenCL AFAIK)?

I like Pi computers, but honestly this is not their place -- yet.

[bruce]

I have no first-hand knowledge of Pi (though my phone uses it.)

What percentage of Pi computers have well-engineered cooling systems and don't depend on battery power. FAH puts a heavy (100%) floating-point based computational load on computers which generates heat and runs down batteries. The Sony client had to contend with that. Laptops (x86) have a somewhat manageable problem with that. My phone would have a problem with that.

Give me a percentage.

[davidcoton]

Pi are mainly mains powered -- 95% or more. ARM chips don't get nearly as hot as AMD/Intel -- partly because they are not so powerful! Pi4B (newest, quickest) uses under 15W including peripherals!

[espressobeanies]

Sorry, it looks like I'm missing out on a good discussion.

Raspberry Pi supports a 64-bit ARM architecture called AARCH64. Fedora Core and Ubuntu Mate both have AARCH64 builds that work with Raspberry Pi 3 and newer models. So Raspberry Pi IS 64-BIT! There are a shedload of Raspberry Pi's out there, otherwise my local Microcenter wouldn't be constantly sold out. The combined computing power of all these Raspberry Pi's would have a "SIGNIFICANT" impact on F@H performance. Why then is Oracle building servers using thousands of Raspberry Pi devices?

Code: Select all

https://www.zdnet.com/article/oracle-this-1060-raspberry-pi-supercomputer-is-worlds-largest-pi-cluster/

To exclude Raspberry Pi devices is somewhat of a disservice. Even though the GPU is not all that powerful, Intel GPUs are pretty terrible when stacked againt NVIDIA and AMD. I'm sure the Raspberry Pi GPU combined with CPU calculating, even though small, spread over millions of devices would add up. So because you don't do it, you will never really know the potential computing power available that Raspberry Pi has to offer. Maybe we need to take this fight to the Raspberry Forums to see if they can dedicate resources to work on a F@H package while we all sit indoors and watch the death toll from COVID-19 accumulate.

[davidcoton]

I don't think the Raspberry Pi Foundation (the not-for-profit organization) or Raspberry Pi Trading (the wholly-owned trading subsidiary) have the resources or the mandate (their registered aims are educational) to take this on.
Consider also the impact on the small F@h development team -- there is quite enough to do without supporting Raspberries.
I do not think a Pi client could be available in time to help fight COVID-19.
While F@h cores need 64-bit (which Pi2B v1.2, Pi3, and Pi4 provide), the base architecture of Raspbian is actually irrelevant.
And finally, try to do some sums on the compute power available. (Forget the Pi GPU, it's totally different from AMD and nVidia.) Look up total Pi sales, work out how many are 64-bit, how many of those might actually be available, and calculate the GFlops available. The compare with the current F@h capacity (somewhere around 2.5 ExaFlops, IIRC). Then decide if supporting Pis is worthwhile.

BTW, I would love it if my scepticism was to prove to be unfounded.