FAH software closed or open?

[ProDigit]

Is the FAH software closed or open source?

By this, I mean to ask, if there was a possibility to make FAH client run any better on my CPU, would I be able to adjust some factors to make this so (eg: enable MMX/SSE or other advanced options; not meaning the 'core count' option of the native client)?

[Joe_H]

The components of the Folding app are a mix of open and closed source - https://foldingathome.org/support/faq/opensource/.

In the case of the computing cores, they are closed source built on tools that are open source. This is because they include digital signing of the WU's and their results to prevent injection of invalid work. In addition, the cores and their processing have to be validated as producing scientifically valid results. Modifications made by end users could not be validated in such a way.

The CPU cores already use SSE2 calculations on all supported hardware, and AVX is used by the A7 core on processors that support that

Portions of FAHClient that deal with downloading and uploading signed WU's from the designated servers are also closed source. Some other portions of the client are open source and posted to Github. I haven't looked at what is posted there in quite a while, so don't recall what portions are available.

You do mention "native client", that is an ambiguous term in connection with Folding@home. It is better to refer to the two different clients for folding on PC's as the "desktop client", current download version 7.5.1; and the Google Na(tive) Cl(ient) for Chrome or Google NaCl.

[ProDigit]

Funny, it's kind of counteractive.
Native kind of means "born of", or "originates from".
I would say the 'desktop client' would run native.
And the native (browser) client, doesn't run natively, because it runs from a browser (even though the browser runs the exe file natively).

Anyway, it appears to me the client is proprietary.
It would be nice if they would open up some part of the program to the public to tune the program to work better with AMD or Intel chips. And from each chip, tune for i3/i5/i7 chips, or Xeons, or Threadripper CPUs.
I think the open source is a good way to get work and improvements done to the core, that will allow better operability, without really modifying the scoring system, or the way the program operates, just to allow it to access parts of the CPU that aren't built in the program, to enhance performance.

Let the open source community take off some of the workload, especially in the Linux community.

[foldy]

I know that the CPU core is based on http://www.gromacs.org/ which is open source. There are 2 different CPU cores A4 and A7. A4 uses SSEx and A7 uses AVX if available. So these should use the CPU resources very good. If anyone wants to improve the CPU cores then best place would be gromacs as the next FAH CPU core will also be based on gromacs.

For GPU core they are based on http://openmm.org/ which is open source. So any improvements for GPU cores would first go into OpenMM and the next FAH GPU core will be based on OpenMM too.

The FahControl UI is open source too https://github.com/FoldingAtHome/fah-control

As this is software used by scientists the release cylces are very long. Which means it can take years until the next core gets released. The reason is they need to verify a long time that a new core really calculates correctly the real biophysics. Because the worst case would be if the work of FAH users gets dumped because the calculations were wrong.

[Joe_H]

Lest you think the worst case doesn't happen, about 10 or so years ago F@h researchers had to restart some GPU projects from the beginning. The reason, a number of folding participants used "hacks" to get the client to fold on some GPU's that had not been whitelisted. The calculations done by those GPU's turned out to be inaccurate and threw off calculations in subsequent WU's.

Current GPU cores include running periodic validation of the WU progress to catch this before it proceeds too far. In addition there are sanity checks done on the returned WU's before they are accepted. All of this is extra overhead, but it can keep the research projects from losing months of work due to invalid results.