Folding Forum

Yes, GPUs.txt is both a whitelist and a blacklist.

I really recommend using https://www.techpowerup.com/gpu-specs/
to check to see if your older graphics can do OpenCL 1.2 or greater and has Double Precision floating point math.

It could still miss deadlines, but at least it could start. (your card looks powerful enough to meet deadlines, if it had DP fp math)

Don't seem to see DP or not specified at e.g. https://www.techpowerup.com/gpu-specs/r ... -5770.c250

But bringing that thing online was basically just for the fun of it anyway; shall be retiring it again.

If there's by the way anyone who could fix the "For technical details" link at viewtopic.php?t=26208 then this would be great. I.e., while it's stated everywhere that GPUs.txt is both a white- and blacklist I can't any information about exactly how it is such anywhere.

Searching for that also made me realise that FAHClient is not opensource. Mmm. I'll leave it up during Corona, but frankly, maybe not much longer...

Yujah wrote:If there's by the way anyone who could fix the "For technical details" link at viewtopic.php?t=26208 then this would be great. I.e., while it's stated everywhere that GPUs.txt is both a white- and blacklist I can't any information about exactly how it is such anywhere.

Searching for that also made me realise that FAHClient is not opensource. Mmm. I'll leave it up during Corona, but frankly, maybe not much longer...

The GPUs.txt list is a list of all GPUs supported by Folding@home. The developers will add more identifiers to the list when new GPUs come out and are proven to work reliably, and they remove them from the list if a particular GPU consistently fails workunits and is unreliable for F@h. The client software will not fetch workunits for a GPU that is not in the list. In that way, it is both a whitelist and a blacklist.

FAHClient is one of the few components in F@h that is not open-source. The original answer, from 20, 15, and 10 years ago still holds true today; F@h is one of the few distributed computing projects that does not duplicate workunits, so it is critical that the software stay proprietary in order to block people from manipulating the client to serve corrupted workunits or cheat the point system. This was a big problem for BOINC and is one of the reasons why BOINC workunits are duplicated to ensure that they are performed correctly, but this is very inefficient. I'm very much in the open-source camp myself, but I can understand this reasoning. The rest of the software is available on GitHub.

Yes, as to the open-source bit, sure, I can sort of see some of the reasoning behind this particular instance possibly being a slighty less standard "security through obscurity" mechanism than is normal for closed-source, and thanks for commenting, and while I wouldn't find myself to be an open-source zealot generally, much less a "free software" one, ...

I've been basically using open-source exlusively for some 20 years now and an "always on" daemon-type application that by design loads up my system significantly but which I can't in fact easily vet myself just doesn't seem to cut it for me anymore. And/or certainly not on actually important or semi-important systems; my current single FAHClient runs on a Windows 7 legacy gaming system, and, yes, sure, go right ahead: that thing's basically my closed-source sandbox anyway, what with the multitude of e.g. decisively closed-source game-stores and games themselves. But on actual systems...

Thanks much for the description of GPUs.txt, although what you describe is then not in fact a "white- and also blacklist" but a standard whitelist. I was assuming that the numeric 3rd and 4th fields would determine status, and especially since as said my Radeon HD 5770 actually is in the current GPUS.txt even though apparently blacklisted server-sides due to no DP.

Again, thanks much for the comment, but that's then still the kind of thing I really just want to be able to check out myself in a source file rather than dig through dead forum links. Oh well...

I completely get your skepticism of closed-source software that you can't inspect yourself. I've been that way myself for a while. But I've also followed F@h for about 8 or 9 years now, and seen how the development happens. It's usually collaborative, with a lot of research teams, PhD students, employees at Nvidia and AMD also contributing their advice and bug fixes. FAHClient is primarily written by J. Coffland from https://cauldrondevelopment.com. I've generally found F@h's style leagues above your average "free download!" software on random websites. I've come to trust the software, have never seen it access or drop files in any unusual ways or make unexpected network requests. It's also been my experience that J. Coffland seems to have a pretty good handle of security too. So I'll endorse the proprietary FAHClient, but of course it's your standards and your call on it.

Note; old information at https://web.archive.org/web/20131022203 ... tOfGPUsTxt (information which would not make explicit e.g. my 5770 not being supported due to no DP whereas a 5830+ is, so how relevant that information still is remains to be seen...)

Old information, they have changed how they define and use cpu_species since that was written years ago.

Now your 5770 would be species 4, no DP support, where the 5800 series are species 5 which does have DP support.