folding@home performance in cloud

Moderators: Site Moderators, FAHC Science Team

toTOW
Site Moderator
Posts: 6296
Joined: Sun Dec 02, 2007 10:38 am
Location: Bordeaux, France
Contact:

Re: folding@home performance in cloud

Post by toTOW »

Also, we measure performance in PPD which is comparable between each computed projects. Time per % depend on the project, so don't trust it.
Image

Folding@Home beta tester since 2002. Folding Forum moderator since July 2008.
foldy
Posts: 2061
Joined: Sat Dec 01, 2012 3:43 pm
Hardware configuration: Folding@Home Client 7.6.13 (1 GPU slots)
Windows 7 64bit
Intel Core i5 2500k@4Ghz
Nvidia gtx 1080ti driver 441

Re: folding@home performance in cloud

Post by foldy »

FahBench will give you results as Score in ns/day which also is comparable.
e.g. GPU gtx 970 has Score 60 with FahBench 2.3.1 on Windows. CPU i5 2500k gets Score 2. http://imgur.com/a/fot5w
Use default settings and just select the Device to switch between different OpenCL GPU devices OR switch Compute to CPU to run the CPU benchmark.

My feeling is you want to explore the Cloud computing thing on your own, would compile your own FahCore if you could and so on. Science is also about experiments so don't hold back. There was a time were FahCore for AMD GPUs was not compiled with optimal settings too and I found the mismatch using FahBench which I compiled myself against AMD SDK and Nvidia SDK (FahBench is open source too).
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: folding@home performance in cloud

Post by bruce »

foldy wrote:FahBench will give you results as Score in ns/day which also is comparable.
It depends on what you mean by comparable.

One minute of computing a protein with 5 000 atoms will produce more simulated ns/day than if those same computing resources are applied to a protein with 500 000 atoms.
PrasadNR
Posts: 5
Joined: Sun Aug 13, 2017 8:31 am

Re: folding@home performance in cloud

Post by PrasadNR »

bruce wrote:While you are seeking to "understand your fatal disease"
Can you let me know if you are offended? I apologise if you are offended for some reason.
bruce wrote: you'll need to understand the difference between FAHClient and FAHCore_xx.FAHClient is essentially a browser. It's function is to download a block of data, wait for it to be analyzed, and upload the results. Even if it is poorly writted (which it may or may not be so what. It contributes almost nothing to the time spent processing FAH. It's job is to provide secure communications with FAH's servers plus monitor when the status of what else is going on and to be guided by some configuration settings.

It is supported by a number of FAHCores which do the scientific processing. They do all the real work, and they are all open source. For CPU processing, check with gromacs.org about the limitation for running with 7 CPU-cores. Check with OpenMM about the code used on GPUs.
I understand that I can try to learn GROMACS. But, I am genuinely interested in making FAHClient better. (I really did not understand how I can get WU from Work unit server and then run it on GROMACS by myself; If I get the source code of FAHClient, I can make this happen possibly)
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: folding@home performance in cloud

Post by bruce »

My point is that the problems you're objecting to are contained in the GROMACS source, which is open. The GROMACS source is not included in the closed source of FAHClient, hence even if you had that source, you couldn't fix the problems you're objecting to.

FAHClient does not process WUs. It invokes a FAHCore_xx which does that and they contain GROMACS (CPU) or OpenMM (GPU).

FAHBench allows you to evaluate stand-alone performance of the FAHCores, and you can download/compile it yourself. FAHClient downloads WUs from the FAH servers. You can run any of those WUs with FAHBench. FAHClient also returns the results to the FAH servers but you cannot do that with FAHBench. This preserves the integrity of the scientific results, preventing hackers from from injecting corrupt or fake results into the scientific datastream.

And No: no offense taken or intended.
dusty
Posts: 1
Joined: Sat Dec 23, 2017 8:46 pm

Re: folding@home performance in cloud

Post by dusty »

I have some free AWS credits and was thinking or running F@H. I'm curious about using the new suite of machine learning tools instead of EC2. My understanding is that the "machine learning" oriented computation that AWS makes use of more GPU processing power, so could be suited to folding. Does anyone have experience with running F@H on those services?
bruce
Posts: 20910
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: folding@home performance in cloud

Post by bruce »

I'm not sure anybody here knows.

Machine learning TOOLS (software) are not useful to FAH. All FAH needs is access to raw computing HARDWARE such as CPUs and GPUs with the appropriate drivers.
foldy
Posts: 2061
Joined: Sat Dec 01, 2012 3:43 pm
Hardware configuration: Folding@Home Client 7.6.13 (1 GPU slots)
Windows 7 64bit
Intel Core i5 2500k@4Ghz
Nvidia gtx 1080ti driver 441

Re: folding@home performance in cloud

Post by foldy »

I know of services like https://vast.ai/ where you can rent a gtx 1080 ti for $0.1/hour as nvidia OpenCL docker image and just install and run FAH with full GPU performance.
Post Reply