COVID-19 Resurface for macOS GPU Support

[shavemadox]

To anyone out there looking to do MORE—

I implore you. I am doing what I can to help fight against Coronavirus. I have downloaded Folding@home through macOS, Windows 10 through Boot Camp and now Ubuntu through VirtualBox. It's not enough. Windows is only convenient to run during the workday, VirtualBox makes me jump through a lot of hoops for little payoff (it still isn't recognizing my GPU) and there simply isn't any GPU support in macOS.

While my Radeon Pro 460 4 GB isn't anywhere near top-of-the-line at this point, I still find it to produce roughly 10x the amount of points using the GPU slot in Windows than the alternative of CPU-only in macOS (TEN TIMES)! I realize that there has possibly never been a busier time for the folks at Folding@home and that Apple may have forced your hand here, but I also feel that it is more important than ever to work directly with Apple to hash out whatever needs to be done to resolve this epidemic as fast as we possibly can.

My understanding is that there is not even any benefit to having an eGPU on macOS with Folding@home—despite having one on the way. I will be forced to also run this through Boot Camp, devastatingly limiting its overall capability. Look, I know us Apple Fan Boys get a bad rap; I realize that overall, we are the weakest link in terms of computing power. I just feel like we should not be punished in a time like this where every little bit helps and that is all we want to do.

Thank you,

Shane Mattox

[PantherX]

Welcome to the F@H Forum shavemadox,

Please note that since Apple had deprecated OpenCL support in favor of METAL, it is unlikely that anything will be done. If Apple were to have an entire team of developers dedicated to getting a FahCore using METAL built, macOS would still not be running live projects on the GPU. Reason being is when a new FahCore is being developed, it goes through lots of internally testing to ensure that the results being produced are scientific value. Historically speaking, this process can take several months. While the "Apple development team" can get that done quicker than a single developer, the bottleneck would be verification of the FahCore results. It would be up to the researchers to analyze the data and see if everything is working as expected or not.

The main issue with the above situation is that while researchers work to validate the results of the new FahCore, it will slow them down in COVID-19 research. Given that Moonshoot is a time-critical research, it is highly unlikely for them to take on another endevour simply because they are already stretched thin. Having said that, if you have Apple contacts who would be keen to develop for F@H, maybe something can be done about it once the pandemic situation is under control.

Regarding your system, please note that while your GPU might generate 10x more points than your CPU, COVID-19 projects that are CPU exclusive needs to be done while others are GPU exclusive. Thus, the scientific importance of CPU contribution is just the same as GPU

[shavemadox]

Thank you so much for the timely response, PantherX! This all makes sense to me; I just wish there were some kind of workaround for the time being. Is it safe to say that any virtual machine is going to perform at a fraction of the rate that it would otherwise? I would really prefer to not have to boot into Windows every time I want to crunch some teraflops.

[Joe_H]

A good VM will perform at close to the full ability of the hardware and OS it is running on as long as it is not starved of resources such as CPU or RAM. There will be some overhead.

Under macOS, CPU folding will be on par with Linux. How well GPU folding in a VM will work I am not certain. All the VMs I have used in the past did not support GPU virtualization well enough to be used. I understand there have been some recent releases of virtualization products that do pass through the access to the GPU hardware, but have not seen any reports on them being used for F@h on macOS.

[shavemadox]

Thank you, Joe_H!!

[PantherX]

It's my understanding (right/wrong) that even if the virtualization supports PCI pass-through (Docker), the reason folding won't work is because the GPU is still in use by the host. The host and guest can't "access" the single GPU simultaneously. Details that I read: https://stackoverflow.com/questions/463 ... radeon-gpu

[Roger.Weihrauch]

Hi to all thread-involved people
Just a hint:
What 'PantherX' said is correct: for VGA/GPU/PCI passing through on a virtualized env (Docker/Vbox/VMM/QEMU&KVM) there will be needed an additional GPU/Graphics card to which the VirtualEnv can have exclusive access to via passing through. In no way there is a possiblility, to my understanding, that the only one GPU (eg. in a MAC) could be (partially/fragmentally) used from a VirtEnv. So, it always needs a second GPU card.
Regards,
Roger