MacOS GPU

[Joe_H]

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?

We could discuss that here, but the discussion that will matter will be among the F@h software developers and researchers. They will need to see enough potential additional systems that can do GPU folding under macOS to justify putting the time and effort into creating the necessary GPU folding cores.

You mention "many MacBook Pro laptops", but only the high end 15" MBP's have discrete GPU's. Among these only the most recent ones are powerful enough. The same applies to iMac's. The iMac Pro is an exception, but they are not a volume system from Apple.

Yes, the OS has OpenCL support, in fact it has had it for many years. But Apple introduced a bug a number years ago around the time of 10.7 the was not fixed until recently. That bug kept early test versions of the GPU cores from functioning properly.

As for CUDA support, is that actually included now? The last I checked CUDA support had to be downloaded as an optional package from nVidia.

[bruce]

Inasmuch as many earlier versions of OS-X have contained corrupt versions of OpenCL, a prerequisite would be an update of FAHClient that would contain code that can detect and report which version of the OS is running. What percentage of Macs have updated to the latest versions?

[foldy]

Another problem for the future is that OpenCL is deprecated in MacOS
https://developer.apple.com/macos/whats-new/
Deprecation of OpenGL and OpenCL
Apps built using OpenGL and OpenCL will continue to run in macOS 10.14, but these legacy technologies are deprecated in macOS 10.14. Games and graphics-intensive apps that use OpenGL should now adopt Metal. Similarly, apps that use OpenCL for computational tasks should now adopt Metal and Metal Performance Shaders.

But you can install Linux as dual boot with MacOS and then folding works for GPU there.

[Theodore]

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?

If it is Intel IGPs, you're in the same boat as windows PCs.
Unless you have AMD IGPs, there's no way to fold on that yet.
I think most research is being done in making dedicated graphics cards working well

[Joe_H]

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?

If it is Intel IGPs, you're in the same boat as windows PCs.
Unless you have AMD IGPs, there's no way to fold on that yet.
I think most research is being done in making dedicated graphics cards working well

The poster explicitly stated MacBook Pro laptops with "discreet GPU hardware". Intel IGP's have no bearing on his post.

[foldy]

Dedicated GPUs do not work under MacOS but only if you dual boot to linux.

[MeeLee]

I wonder if you can fold from GPU on a mac, from a virtual machine?

[toTOW]

And with Apple plans to move from x86 architecture to ARM one for their desktop and laptop machines, GPU support is very unlikely. We will probably even loose CPU Folding support with this move ...

[MeeLee]

And with Apple plans to move from x86 architecture to ARM one for their desktop and laptop machines, GPU support is very unlikely. We will probably even loose CPU Folding support with this move ...

But you'll gain a wonderful Apple system that's much more energy efficient!

The move from x86 to ARM has been a long standing topic of discussion.
Many decades.
The main reason it was never implemented, was the lack of performance of the ARM chipset at the time, and software support;
But now with mobile phones all running android, at speeds nearly 10x of what it was 5 years ago, it no longer is an issue.
Nearly every (home user, non-professional) desktop program out there used today in a Windows environment, has an equivalent on the ARM chipset. (think browsers, Office packages, games//steam, apps and accessories, system tools and more...).

If they would port Apple OS to ARM chipsets for laptops, I think it would be a smart move on their side; just for the fact that ARM chips are more efficient than x86 chips, and Unix has been proven to run reliably on ARM.

I believe Windows is heading towards the same direction as well. They're also porting Windows to ARM.


x86 is dying. We all knew the old cow would give up someday, and I don't think it would take much longer.
I'd be surprised if Microsoft would make more than just one last Windows upgrade on x86.

The way things are going, the gap between GPU and CPU folding is getting wider and wider.
It surprises me a lot of people are still using CPUs for folding.

Perhaps I'm secretly hoping for some company (Nvidia?) to develop PCIE cards just with processing cores on (like stripped down graphics cards), for deep learning or AI development.
Something like that would have great potential for folding; but would probably need an entire rebuild of software to fold on that.

[bruce]

Perhaps I'm secretly hoping for some company (Nvidia?) to develop PCIE cards just with processing cores on (like stripped down graphics cards), for deep learning or AI development.
Something like that would have great potential for folding; but would probably need an entire rebuild of software to fold on that.

It should be noted that this device wouldn't run MacOS or Winodws or Linux ... it would just be a math co-processor, supported by drivers, You'd still need an operating system running on some other device so the market for that device would be small. Adding back in hardware to rasterize and display images would allow it to be sold to a wider audience ... and then you'd have what we have today.

[toTOW]

And with Apple plans to move from x86 architecture to ARM one for their desktop and laptop machines, GPU support is very unlikely. We will probably even loose CPU Folding support with this move ...

But you'll gain a wonderful Apple system that's much more energy efficient!

But that's not what we're seeking here ...

And ARM CPUs are crap for computing ... I'm not certain that they could even feed a big GPU efficiently ...

[bruce]

You're both right ... but about things at opposite ends of a spectrum.

if you're objective is to run FAH, you want to maximize the computational throughput which is directly associated with a high power demand. If your objective is to have a fast browser that lets you search the internet or to watch a video without high compute processing capabilities, you can maximize energy efficiency, which is great for a battery powered device (laptop or phone).

Basically, you've got 4 choices. A low power ARM for a phone, an x86 optimized for low-power x86 for a laptop, and a high power device for FAH performance, with an x86 optimized for high-computational loads, followed by a discrete GPU for maximum computational throughput and power demands.

Note that the FAH client for the ARM client on a phone has NOT been a big winner and it probably never will be because its minimal computational capabilities.

You have to choose between power efficiency and computational throughput.

The ARM is probably capable of moving data across the PCIe bus but recent OpenMM developments have migrated increasing computational loads to the x86 (or more exactly, the SSE2 processor) increasing the total throughput. If Apple migrates OS-X to ARM and deprecates OpenCL, that may indicate an upcoming demise of a FAH client for the Mac.

[MeeLee]

Yeah, I wouldn't recommend to fold directly on ARM chipsets either,
But, I do think they could provide enough throughput for discrete graphics cards.
For example, Snapdragon 845 has a 2,8Ghz CPU. It should be enough to feed an RTX2060.
Fact that it has 8 of them in a package, there is potential for multi GPU systems.

I guess you're right, that a computational device like mentioned above, would still need graphics memory; which would make it a graphics card.
I've read somewhere, that 50% of current cost of graphics cards, is the memory.
It would be nice to have a graphics card with less memory, and more cores. Eg: An RTX2080 ti with only 2 to 3GB of RAM could cut down the cost of such unit considerably.
Chances on that happening is small, as it wouldn't translate into a great gaming card, so I don't see Nvidia starting up a market for it.

Then again, GDDR6 is relatively new; and ram prices will fall, like with any technology; so prices on high end cards will go down over time.

[bruce]

I may be wrong, but if I have a MacOS system and want to install a GPU (say an RTX2060) somebody would have to supply the graphics driver for the TU106-200A-KA-A1 chip that can be installed in OS-X, running on either an x86 CPU or an ARM CPU. Do such drivers exist?

[MeeLee]

Not sure how MacOs runs Windows, and if Windows drivers would work from an emulation perspective.
There are some sites where users of Macs are running windows games through some sort of emulation, like Wine is to Linux.
I don't have a Mac.