Folding Forum

OSX support would be interesting

foldy wrote:OSX support would be interesting

That I see getting less likely with each passing month. Apple has moved to using discrete GPU's from nVidia and AMD only in the Mac Pro and high end 15" MacBook Pro's and iMac's, that leaves such a small group of systems to target that creating GPU folding cores may not make sense. The integrated Intel video in the rest of the systems sold has gotten more powerful, but there is no sign that the OpenCL support from Intel is ever going to be sufficient for folding use.

Joe_H wrote:The integrated Intel video in the rest of the systems sold has gotten more powerful, but there is no sign that the OpenCL support from Intel is ever going to be sufficient for folding use.

I want to add emphasis to that statement. The Intel integrated GPU is not supported on Windows/Linux/OSX. It's just too slow to make it worthwhile.

Integrated GPUs from AMD are in the same category. Some of the recent ones do provide supportable drivers, but they are also too slow to be worthwhile. Moreover, they'll complete with the AVX core for the same resources.

Question: How many of you with OSX actually have a discrete GPU from AMD or nVidia? There would need to be quite a few of you for Stanford to port the FAHCore to OSX and develop the infrastructure to send WUs to those people.

Since Apple fixed the OpenCL bugs it should be easy to build the FahCore also for OSX and the infrastructure is already there. But if there are other bugs showing up it is the question if it is is worth tracking them down. The Standford internal statistics should show how many OSX have also AMD or Nvidia GPUs. The public statistics show 8351 active CPUs and if we assume only 1 out of 8 have an AMD or Nvidia GPU this is still 1000 GPUs ready for folding.

What kind of "internal statistics" do you think Stanford has?

From what I know, every client reports its configuration when it requests a new assignment and since OSX clients have no supported GPUs, nothing is reported that can be used. The 8351 active CPUs is based on history ... WUs which have been completed.

A list of recently completed WUs can be scanned for the type of hardware used to complete them, but that's not useful since (obviously) none have been completed by OSX GPUs that might be supportable.

That's why I asked folks with NV/AMD discrete GPUs on OSX that might be supportable to check in.

Just as a data point for example, I have three Mac's folding. One is a MacBook Pro with an integrated Intel GPU, another is an older iMac with a Radeon 4850M GPU, and finally my son's Mac Pro has a GTX 680 as a replacement for its original HD 5870. So one potential GPU folder under OS X.

I believe many with an old Mac Pro and recent GPU are running Windows on it.
True for me anyway.

I would get an older MP if GPU works; so I have to use Linux ... I just hope that at least CPU and FAHContril/Viewer remain supported

calxalot wrote:I believe many with an old Mac Pro and recent GPU are running Windows on it.
True for me anyway.

Some are running OS X still, Apple included support in later versions starting with 10.8 or 10.9 for use of standard PC video cards instead of Mac specific models. The main downside is that there is no video during boot up until the driver loads as part of the OS. That is what my son is doing, the GTX 680 he got from a friend who was upgrading his PC.

bruce wrote:What kind of "internal statistics" do you think Stanford has?

The hardware config which a client reports(CPUs and GPUs)? If that is not available we can only guess.

Hardware on slot, config (like slot-size) and resulting TPF and OS ...
At least for GPUs that would be great.

I don't think they store that data anywhere. Donors supply whatever they have and FAH does it best to generate science from it. A DB of that information wouldn't generate more science.

i.e.-That is not available. We can only guess.