Folding Forum

http://www.techreport.com/discussions.x/14908
Is there a hope for GPU2 client for MacOS X?

We saw the announcements too, but no reason to expect anything.

Breath holding not recommended.

There will likely be CUDA and CAL for OSX, so we'll use those. My plan is to get the Windows versions running well before thinking about Linux & OSX.

Please make a Linux client at some point so i can fold GPU2 workunits on these machines. For now its SMP/Uni only.

v00d00 wrote:Please make a Linux client at some point so i can fold GPU2 workunits on these machines. For now its SMP/Uni only.

I'll add my vote for Linux GPU2 client as well.

same here.
I gave too much of my lunch money to Bill already.

I've been reading up on CUDA (with the new NVIDIA workstation cards, I had to check it out) and it seems like OpenCL IS CUDA in some sort of wrapper*. The difference between CUDA and OpenCL is that OpenCL is cross-platform. (i.e. NVIDA, ATI and Intel GPUs)

Of course we won't know all of this until 10.6 comes out

http://forums.nvidia.com/index.php?showtopic=69731

VijayPande wrote:There will likely be CUDA and CAL for OSX, so we'll use those. My plan is to get the Windows versions running well before thinking about Linux & OSX.

That's understandable because thats where the majority of folders are, but i think Snow Leopard's new set of technologies could really spur on a lot more Mac users to fold. I think a Mac client with longer deadlines needs to be brought out in the meantime. It would really help Mac users get on board without having to leave their systems on 24/7

I think the main issue on the Mac side is that people are just not buying the Mac Pro desktops we need for folding anymore, they buy laptops and iPhones. And I doubt the iPhone will be a great 24/7 folding platform any time soon

Beberg wrote:I think the main issue on the Mac side is that people are just not buying the Mac Pro desktops we need for folding anymore, they buy laptops and iPhones. And I doubt the iPhone will be a great 24/7 folding platform any time soon

No reason why people can't and don't fold on MacBooks.. i know i do!
Please don't be silly, of course the iPhone won't be folding. All i'm saying is that the Mac deserves to have a super simple client and an SMP client or something that can atleast let you set what sort of deadlines would be acceptable with your usage etc.

smASHer88 wrote:

Beberg wrote:I think the main issue on the Mac side is that people are just not buying the Mac Pro desktops we need for folding anymore, they buy laptops and iPhones. And I doubt the iPhone will be a great 24/7 folding platform any time soon

No reason why people can't and don't fold on MacBooks.. i know i do!
Please don't be silly, of course the iPhone won't be folding. All i'm saying is that the Mac deserves to have a super simple client and an SMP client or something that can atleast let you set what sort of deadlines would be acceptable with your usage etc.

I don't disagree, but when the SMP client was developed it was defined as a "high performance" client and the deadlines were intentionally set to be consistent with multiprocessor systems. Surely you don't expect that the Mac should have special deadlines on their WUs that are unavailable to Linux or Windows users. I have a 2.8 GHz P4 which has one core and HyperThreading so it acts like it has two cores. The SMP WUs will run on on, but they take twice to four times longer than they would if it were a real multicore system. Stanford does NOT want me running SMP on that system because it would slow down the SMP projects.

The real issue here is that the projects assigned to the uniprocessor client have deadlines that can easily be met by almost any computer, but unfortunately the uniprocessor client for the Mac run on PPC, not Intel, and Rosetta is a very poor option. I have asked the Pande Group to develop some kind of client that is equivalent to the v5 client which works effectively on Intel Macs but they have not and I have no way of knowing if they ever will.

Does Parallels or some other virtual system allow you to run the Windows uniprocessor client? Is there a rational way to run the Linux uniprocessor client? Either one would allow you to process WUs with much longer deadlines on machines which only operate part of the day. None of these solutions meets your "simple" requirement, but perhaps a 3rd party developed shell could make it appear simple.

I would think the windows uniprocessor client would run under wine for mac.

I think Beberg was responding playfully about the iPhone...

Many people fold SMP units successfully on Intel Mac laptops. I suspect the Pande Group is asking themselves if there are enough Intel Mac desktops (Mac Pro's, which can take graphics cards up to 300W total power demand) to warrant developing a GPU client for Mac.
Currently the GPU's that come with the Mac Pro are either a NVidia GT 120 (a low power OEM-only card with 32 processors), or ATI Radeon 4870. The NVidia 8800GT "For Mac" in my 2008-vintage Mac Pro is not compatible with the newer Nehalem-based Mac Pros.

I had been advocating a "GPU on wine" approach for OSX, but since the newer Mac Pros use either an ATI card or fairly low powered NVidia cards, this approach would only be useful for the older MP's. AFAIK GPU on wine is limited to NVidia cards (counting out the 4870) and the 32-processor GT 120's don't look powerful enough to be worth folding with. Anyone have an idea what 8000 or 9000 class NVidia card is comparable to the GT 120?

susato wrote:Anyone have an idea what 8000 or 9000 class NVidia card is comparable to the GT 120?

Chart found here for comparison. Based on this announcement in late 2008.

Consumer rebranding;
G100 = 9400 GT
GT120 = 9500 GT
GT130 = 9600 GSO = 8800 GS

OpenCL for non-MAC is coming:
http://www.amdzone.com/index.php/compon ... st-opencl-