eGPU Developer Kit incompatible?

[LightBulbFun]

HEVC encode and decode support for Vega GPUs is shown in the AMDRadeonX5000.kext driver.

View attachment 742634

im glad someone finally mentioned this

I also believe the hackintosh guys have been able to add these strings to the Polaris kexts and get HEVC on the go there as well but dont hold me to that

I do wonder very much if Vega HEVC decoding/encoding works with a Vega Card in a 5,1? or if Apple is only allowing it on the iMacPro1,1 somehow currently...

 

[MisterAndrew]

im glad someone finally mentioned this

I also believe the hackintosh guys have been able to add these strings to the Polaris kexts and get HEVC on the go there as well but dont hold me to that

I do wonder very much if Vega HEVC decoding/encoding works with a Vega Card in a 5,1? or if Apple is only allowing it on the iMacPro1,1 somehow currently...

I suppose I could test it, but there’s no indication Vega HEVC wouldn’t work on the cMP. The driver lists the device ID for the card.

 

[deconstruct60]

It's not. There are Thunderbolt 2 machines not on that list.

The Nvidia ones are there because Apple doesn't support AMD and Nvidia on the same box.

Intel GPUs and Nvidia/AMD mix, but the AMD and Nvidia mixing blows up the graphics stack somehow?
That's odd and smells as if this interim eGPU support is a more than a bit of a kludge just happens to work as opposed to a rational, well designed graphics stack.

 

[goMac]

Intel GPUs and Nvidia/AMD mix, but the AMD and Nvidia mixing blows up the graphics stack somehow?
That's odd and smells as if this interim eGPU support is a more than a bit of a kludge just happens to work as opposed to a rational, well designed graphics stack.

Intel and Nvidia GPUs never both render screen output at the same time.

And I agree it's not great, I think a lot of the breakage we've seen in other stuff in graphics is them trying to fix it but breaking things at the same time.

 

[deconstruct60]

Intel and Nvidia GPUs never both render screen output at the same time.

In a eGPU setting why not? The context here is an eGPU situation. All the MBP's will have the display hooked to the internal GPU(s) snd it is probably on. For a MBP 13" that pretty much has to be an Intel GPU which is active. Even if there is an internal GPU and some soft/hard DP switch between those two what in the world would have to do with coupling to a second GPU on a different bus in any sanely designed system.

And I agree it's not great, I think a lot of the breakage we've seen in other stuff in graphics is them trying to fix it but breaking things at the same time.

Which all the more sounds like the graphics stack is effectively held together with a mass of spit , bailing wire, and duct tape. (either that or have had significant talent bleed off and few, if any, have deep institutional knowledge of the codebase. ) It isn't surprising that Apple has ducked trying to do eGPUs for so long. They'd probably still be ducking if could get away with it.

 

[goMac]

In a eGPU setting why not? The context here is an eGPU situation. All the MBP's will have the display hooked to the internal GPU(s) snd it is probably on. For a MBP 13" that pretty much has to be an Intel GPU which is active. Even if there is an internal GPU and some soft/hard DP switch between those two what in the world would have to do with coupling to a second GPU on a different bus in any sanely designed system.

Oh right, I forgot some machines don't have a dGPU. I dunno then, Intel must not have the same problems.

I think unlike the Nvidia and AMD drivers, Apple writes the Intel drivers which gives them a lot more control. Nvidia and AMD talking through Apple on a delay is part of the problem.

Which all the more sounds like the graphics stack is effectively held together with a mass of spit , bailing wire, and duct tape. (either that or have had significant talent bleed off and few, if any, have deep institutional knowledge of the codebase. ) It isn't surprising that Apple has ducked trying to do eGPUs for so long. They'd probably still be ducking if could get away with it.

There is no real architecture. It's why GPU drivers still can break between macOS versions, and the GPU driver interface isn't even documented.

(An actual architecture might fix the AMD and Nvidia don't talk well problem I mentioned above.)

With eGPU there is going to be a lot of pressure to fix that.

But, to be fair to Apple, Nvidia and AMD on the same Windows box wasn't officially supported for a long time either. I think it is now.

 

[jasminetroll]

It does. There was a report of dual RX 580 with a D700s nMP.

Sort of, at least in the case of my D300 6,1 running 10.13.2. The (Sapphire Pulse 8 GB) RX 580 only works correctly in the eGFX as a compute card; connecting a display causes a hard reset within seconds.

OTOH, NVIDIA Pascal (MSI 1070 Ti tested) works for both display and compute, in this same configuration, using NVIDIA’s Web drivers and
https://egpu.io/forums/mac-setup/wip-nvidia-egpu-support-for-high-sierra/
so the problem is most likely with the (Apple-supplied) AMD drivers, rather than with any of the hardware.

 

[MarkJames68]

Sort of, at least in the case of my D300 6,1 running 10.13.2. The (Sapphire Pulse 8 GB) RX 580 only works correctly in the eGFX as a compute card; connecting a display causes a hard reset within seconds.

OTOH, NVIDIA Pascal (MSI 1070 Ti tested) works for both display and compute, in this same configuration, using NVIDIA’s Web drivers and
https://egpu.io/forums/mac-setup/wip-nvidia-egpu-support-for-high-sierra/
so the problem is most likely with the (Apple-supplied) AMD drivers, rather than with any of the hardware.

And I have the XFX which is usually the kiss of death - yet works as a display card to a 21:9 1080 monitor.