GPU still not supported for folding on MacOS client?

[sfobrian]

Hi there,

I rejoined the community after a very extended break to help with the COVID-19 effort. In setting up on a MacBook Pro with a (whitelisted) GeForce GT 750M Mac Edition, I'm unable to get the GPU to download a work unit even though the CPU slot is working fine (albeit intermittently due to the work unit shortage). In looking at the log (snippet below), it appears that the targeted server it's trying to download from is the problem.

Seeing as it's trying to connect to a reserved 192.0.* IP in order to download the work unit, well, that's never going to happen. Is the 192.0.* targeted IP a bug, or is this just a very roundabout way of saying that GPUs are still unsupported in the Mac client?

Searching around on the forums, I'm finding posts from a few years ago implying things were going in that direction – but nothing within the past year or so (or a clearly definitive statement regardless of year). If I can't put my GPU to use on this, that's fine (although disappointing as it's just sitting idle) – I'd just like to know if there's a fix or I'm wasting my time hunting for one.

Thanks in advance,
Brian

Code: Select all

******************************* Date: 2020-03-19 *******************************
13:36:44:WU01:FS01:Connecting to 65.254.110.245:8080
13:36:44:WU01:FS01:Assigned to work server 192.0.2.1
13:36:46:WU01:FS01:Requesting new work unit for slot 01: READY gpu:0:GK107M [GeForce GT 750M Mac Edition] from 192.0.2.1
13:36:46:WU01:FS01:Connecting to 192.0.2.1:8080
13:38:02:WARNING:WU01:FS01:WorkServer connection failed on port 8080 trying 80
13:38:02:WU01:FS01:Connecting to 192.0.2.1:80
13:39:17:ERROR:WU01:FS01:Exception: Failed to connect to 192.0.2.1:80: Operation timed out
******************************* Date: 2020-03-19 *******************************
19:36:47:WU01:FS01:Connecting to 65.254.110.245:8080
19:38:41:WU01:FS01:Assigned to work server 192.0.2.1
19:38:41:WU01:FS01:Requesting new work unit for slot 01: READY gpu:0:GK107M [GeForce GT 750M Mac Edition] from 192.0.2.1
19:38:41:WU01:FS01:Connecting to 192.0.2.1:8080
19:39:56:WARNING:WU01:FS01:WorkServer connection failed on port 8080 trying 80
19:39:56:WU01:FS01:Connecting to 192.0.2.1:80
19:42:21:ERROR:WU01:FS01:Exception: Failed to connect to 192.0.2.1:80: Operation timed out

[Joe_H]

GPU folding is still not available on the macOS. There have been some recent questions about that, but depending on the search they may not show right up.

The redirect is to a specific reserved IP range for test and documentation, a future version of the client may display a message that the request can't be filled when that version finishes development. In the meantime it keeps the request from adding further load on he Work Servers. Requests from a variety of no longer supported and unsupported configurations get the same kind of redirect.

Currently the only way to fold using a Mac's GPU is setting up an install of Linux or Windows.

[sfobrian]

Thanks Joe. Yeah, in continuing to try to root out an answer in the time you took to respond (so quickly - thanks!), I saw your response on the same topic in a different sub-forum from a little over a year ago. (Sorry my search-fu was a little lacking before posting)

It's a bummer, but I totally understand the tough job of prioritization of development work. I do my alternative OS'ing in Parallels, not Bootcamp so my CPU will have to be enough.

Thanks,
Brian

[marook]

Same issue here.
Sad, that we can't add GPU's to the workforce.. :-/

[Joe_H]

Yeah, years ago when they first started development of the v7 folding client, the intent was there to include OS X for GPU folding. But around the time the first of the OpenCL GPU folding cores was being developed they found that Apple had introduced a bug in their OpenCL support, and that was a showstopper. Bug reports were filed with Apple, but it was a couple years or so before a fix was released.

Sort of a shame that Apple took so long to fix a bug in a standard that originated with them, by the time it was fixed there were very few usable GPUs in the current and prior models of Mac's. The inertia of already having a working core for OS X might have carried it through that period if Open CL had been working before on OS X. Now that there are some usable GPU chips showing up in the model lineup I understand there is some interest, but Apple has deprecated OpenCL in favor of Metal. So there is uncertainty on how long a core developed for Mac's would be usable.

[EnderTunin]

Would be nice to have the Radeon Pro Vega 64 of my iMac Pro running these. With so many joining lately maybe now the impact the mac users would have would be worth it?

[bruce]

I'm afraid that if you eliminate the Macs that run on an integrated video system, the total number of actual GPUs is pretty small. Does anybody know the statistics?

[jonault]

Most desktop Macs have discrete GPUs - the only ones that don't are the Mac Minis.

Most laptop Macs have integrated graphics - the only ones that have discrete GPUs are the top end MacBook Pros (currently the 16" models, prior to that the 15").

Apple doesn't release specific sales numbers anymore, but when they did they showed Mac laptop sales outnumbering desktop sales; I doubt that's changed. So I'd say the majority of Macs out there wouldn't support GPU folding even if it was available.

Personally, even though I have a 16" MacBook Pro that could fold on its GPU, I doubt I would do it - it's my main general purpose computer & I use it all over the house, so having it permanently tethered to a power supply so it could fold would not be appealing; I don't even do CPU folding on that one. The only Mac I fold on is the stationary one (a Mac Mini).

[seegem]

Hi Everyone,

I have some experience with this. I'm a Mac user (exclusively, always have been), and for the last 20 years I've devoted my spare compute time to SETI @ Home (searching radio signals from space for extraterrestrial intelligence). The same issue arose that GPU compute on Macs wasn't supported, while on Windoze it was. I hired a programmer at my own cost to fix this ~3 years ago, and every since SETI@Home supports CPU & GPU compute on Mac OSX. SETI@Home is now retired, and I'm switching to Folding@Home. Given the number of GPUs out there in the Mac world, I would strongly encourage someone with the skills to do the work to get GPU compute on the mac platform to work. The fleet compute performance jumped over 30% enabling GPU support. That's huge.

Cheers,

Mark

[ipkh]

It's a political matter within Apple itself that denies Folding At Home the use of Mac GPUs.
They deprecated OpenCL in favor of their Metal API to the point where it is no longer compatible with FAH.
They no longer sign manufacturer drivers preferring to keep things in house.

[toTOW]

You can still boot your Mac on Windows or Linux and use the GPU there.

[durval]

As per my comment on

Code: Select all

github: github.com/openmm/openmm/issues/2489#issuecomment-612541232

Folks, there's something wrong; according to Folding@Home's own "OS Stats" page, there are exactly 65 MacOSX active AMD GPUs right now:



How come?!

Whatever it is that these folks are doing, I want in too: I sysadmin a MacOSX server with two very capable AMD GPUs that are completely idle because the machine is headless (ie, has no keyboard/mouse/display) and is only used remotely via SSH... would love to contribute this otherwise wasted resource to the F@H COVID-19 effort. And I know some other people in the same situation...

Thanks in advance to anyone who can help shed a light on this!

Cheers,
-- Durval.

[PantherX]

Please note that it is a copy/past issue made by the researches while setting up the project descriptions. It is known and will be fixed eventually. There's no GPU support for macOS currently.

[durval]

Hello @PantherX,

Thanks for your response! More questions/comments below:

Please note that it is a copy/past issue made by the researches while setting up the project descriptions.

Uh?! How so? Aren't these stats calculated automatically?

It is known and will be fixed eventually.

I've been seeing this number change, and not to zero; for comparison's sake, here's the latest:


So, how come "they will be fixed eventually"? Are they somehow being fixed peacemeal, ie, just some records every few days? Isn't that strange?

There's no GPU support for macOS currently.

OK, but what about previous versions? Perhaps some"closed" (ie, not public available) beta version got used some months ago, ran for a while, and is now its stats are being slowly purged by the normal "not active for the past 50 days" procedure?

Cheers,
-- Durval.

[Joe_H]

The number was 0 in that field until recently, no Beta tests, no internal tests, to make it anything other than zero.

And yes, they have to fix it piecemeal. The origin of the error comes from a number of related projects where some aspects were being set up for investigation through CPU based projects and WUs, and other aspects were to be GPU based. So copy and paste descriptions and other bits, and not notice a "C" has not been changed into a "G" or vice versa. So now they have to go back and fix individual locations where this happened and propagate the information through the work done while the setting was misidentifying CPU versus GPU.