Folding Forum

Hi guys!

Are there any plans to make a MacOS GPU client? I have 3 x 980 Ti and would love to use them.
-
Dave

If you use linux on Mac then it is possible.

What Mac can deal with 3 GTX 980 Ti ?

In principle I wish too; at least with single GPUs; even more as the OpenCL issue is solved; but seems the share of capable Mac is too low -> as foldy suggested: use Linux ... I do with CentOS until macOS one far far day in the future might get GPU support under FAH.

It has been discussed (such as here.... Nothing official that could be called "plans" though.
and
Like ChristianVirtual says, the fundamental problem is that most MacOS systems use embedded GPUs which are too weak to add significant productivity to FAH. Systems like yours with powerful GPUs are the exception so the efforts/costs associated with adding support for such systems would be less productive that spending scarce development efforts elsewhere. Debugging anything new like that capability isn't a trivial undertaking.

Thanks for the Linux advice. And yes its a real shame Apple has totally lost interest in high end GPUs, there is such little incentive for developers to support OSX.

I need tons of CUDA performance for 3D rendering so I just build PCs and run MacOS. Thats how I got the 3 GPUs in there.

I could just switch to Windows or Linux but since I run a studio with a ton of Macs I don't want to start mixing OSes just for FAH.

The lack of Mac GPU support is a shame because the Mac Pros have some OK GPUs...even if they are SOOOOO overpriced for 2017.

Apple's use of discrete GPU's has changed over the years, unfortunately in a direction that makes it less likely that there will be an OS X GPU folding core. The original intent of the V7 client going back 5-6 years or more was to include OS X in GPU folding. Then a number of models included discrete GPU's, though all but the Mac Pro were mobile GPU based implementations. Since then all but the 27" iMac and 15" MacBook Pro models have gone to integrated GPU's outside of the Mac Pro line. The FirePro based GPU's used in them are relatively powerful, but a 3+ year old GPU.

What I take from this is that a small fraction of the OS X machines have usable GPU's, perhaps not big enough for PG to expend development time to create a folding core and test it sufficiently for production use. They can not base their decision making on an unknowable number of Hackintoshes running OS X and created outside the officially supported machines from Apple, or 5+ year old Mac Pro models upgraded with GPU's originally meant for PC's. My personal opinion is all of this makes it unlikely that a GPU core will be created for OS X unless the hardware utilized by Apple changes.

I say this as an OS X user almost exclusively at home. The one bright point lately has been the A7 core for CPU folding, it definitely gets more performance out of my OS X systems, especially my newer MacBook Pro which has a CPU that supports AVX.

If I look around my Mac's there is only a 2010 27" iMac i would fold with. My MBP get too hot and the MBA/Minis are too weak plus heat again (NaCL from time to time was running).
The only serious candidate in my eyes was the former Mac Pro with slots and room for GPUs.

This said: I hope and pray that the FAHViewer and FAHControl will long be supported on macOS. Because that's what I most need.

First there were x86 CPUs. Then came 3DNow and SSE which was a major performance improvement. Then came Discrete GPUs which was another big step forward.

For a long time 99% of the donor world shifted to SSE and dGPUs. GPUs have gone through multiple generations creating a large gap between what can be done with most CPUs and most GPUs. Thankfully, AVX now is augmenting what might have been assigned to SSE, filling in part of that gap . I wonder what hardware will provide the next generational shift.

I'd like to challenge a few points here.

* "[Most new Macs don't have GPUs]"
** Many higher-end "Pro" Macs and MacBooks do contain discrete GPUs like those from Nvidia or AMD.
** Even without true discrete Nvidia or AMD GPU's, lower end models have Intel GPU cores either in the chipset or CPU. Are these of no utility?

* "[PG see's too few Mac OS X users with GPUs to make support a priority]"
** That may be true, but would they be opposed to open sourcing the work to those that care, have the time and the skills to research and add support?

-NGMFAH

As a fraction of sales the high end Mac's with discrete GPU's are a small portion of all that Apple sells, so the first statement remains true. As I stated in my earlier response that could change if Apple changes what GPU hardware gets used.

The current models of Mac Pros have a nearly 5 year old tech as GPU's. Usable but not a large number of systems. As for the level of discrete GPU used in the high end iMac's and MacBook Pros, that has been tested as mentioned here - viewtopic.php?f=104&t=29327&start=15#p291095.

Intel GPU cores are not usable for folding, and that applies to OS X, Linux and Windows.

As for support, PG reevaluates that as conditions change. The linked post includes one such evaluation. However the folding core is unlikely to be open sourced.

Hello All,

Just to let you know that many MACPRO users have current high end GPUs running on their machines and that iMacs Pro now
have also very powerfull GPUs. I run many of them on different 3.46GHZ Xeon 24 cores machines, including AMDs RX580,
NVIDIA 1070ti, and a 1080tis ...

I really dont understand what makes it so difficult for your wonderful (yes !) team to support OpenCL or CUDA on OSX/Darwin,
or ask Apple to help ! Considering current Apple directions I would start with an OpenCL / AMD implementation...
As you know these GPUs provide much more compute power than the Xeons processors.

Hope you will find some resources to implement that. I would be willing to beta test if you need some help.
Best regards. Serger1162.

Welcome to the folding support forum Serger1162.

First your definition of "many" may not match what the developers of the folding software would consider sufficient to devote time and other resources to creating a folding core and maintaining it for OS X. From experience it represents a small fraction of the Mac Pros that Apple sold. Your examples of GPUs installed for instance are all non-original and not directly supported by Apple. I also have to assume the "3.46GHZ Xeon 24 core" machines are upgrades, and that you mean dual hex-core CPU's with 12 cores total and 24 threads. These are upgraded 6+ year old machines as well, how many will stay in use going forward?

Driver support for the nVidia cards with Maxwell and Pascal chips is supplied by nVidia as none were used by Apple as OEM video, and I can find CUDA support but not OpenCL support downloads on their site. It is possible that OpenCL built into macOS is sufficient, but that would have to be tested with these GPU's. Apple does support OpenCL computation on AMD based GPUs. Driver support is good for specific GPUs from AMD used by Apple, but varies in how well it works for related GPUs from AMD not utilized by Apple. The PG developers have to depend on the OpenCL support from the GPU makers and the OS, they do not provide that.

The newly released iMac Pro do look promising as the first iMac models that do not use mobile GPU chips for their video. They also have an improved cooling system that may be sufficient to keep them cooled enough to run at a large fraction of their capability continuously. Overheating of the GPU has been an issue in prior iMac models, leading to shortened lifespans. With their Vega based GPUs, they have ones that are good matches to those available to Windows and Linux users.

OK you seem to know what you are doing however I would like to understand your definition of "many".

What do you think about the fact that in about two months of light calculations, i.e. only sparsely running my Macs
under windows & bootcamp over night I already have the overall rank 39810 of 1908210 ...

I fully understand your investment in the 1 868 400 users which have contributed a lot however sometimes I feel,
as many Macintosh and Hackintosh users that there is a biais towards some operating systems because they appear
easier to developpers than OSX / Darwin and sometimes more "fashionable".

On your other point, regarding OpenCL support for Linux/MAC from NVIDIA it is finally available. Problem is solved.
Please take a look here.
https://developer.nvidia.com/opencl

Regarding mobile GPU chips i dont understand your statement. They are powefull compared to CPUs, but are just designed
to use less power than the full fledged ones.

Anyways thanks for your reply, keep the good work going, and if at some point you change your mind (thanks to the new macpro) dont hesitate to
call for help or beta testers...

Historically, many older versions of MacOS X officially contained OpenCL support but it did not meet the minimum requirements to run FAH's Cores. (Yes, FAH was thoroughly tested.) A blanket prohibition was the right decision. (As Joe has said, hardware support has also been rather limited.)

Recently, Apple has been making the kind of changes required and personally I do agree that it's time for FAH to reevaluate the current state of affairs -- though FAH has made no official statement one way or another. FAH is interested in simultaneously making the best decisions possible about where to apply their limited development funds so as to maximize overall production. Joe's use of the word "many" is suggests that a realistic tradeoff needs to be made before undertaking a new project.

An important question that would also need to be considered is how should the servers assign WUs that can be completed while avoiding situations where the WU will be dumped because that particular machine will not successfully complete the assignment promptly.

Support IS BEING ( will be re-) CONSIDRED but it's not as simple as you make it sound and the development backlog is long. Getting to the top of that list is a real challenge given the number of good ideas compared to the resources available at FAH.

Given that MacOS now has OpenCL and CUDA support built-in, and given that there are many MacBook Pro laptops with discreet GPU hardware capable of out-performing the CPUs in those laptops, I'd like to propose that we re-open this discussion?