HVEC on iMac Pro update

[Trebuin]

Apple updated their iMac Pro info page and includes a lot of not so clear benchmarks. One of these struck my eye referencing using the Intel Media Engine and I'm wondering if they clocked QS on it. My first though is "absolutely not" because the Xeon would need QS support, which would mean an intel graphics card on chip...which the Xeon should not have. All I can say is time will tell, but if these have it on it as a custom chip...well then. (I still doubt it)

 

[joema2]

Apple updated their iMac Pro info page...the Intel Media Engine and I'm wondering if they clocked QS on it. My first though is "absolutely not" because the Xeon would need QS support, which would mean an intel graphics card on chip...which the Xeon should not have....

That is a good question. There are rumors the Xeon-W "B" variant used in the iMac Pro has been customized with Quick Sync, but nobody knows for sure. However the limited benchmarks involving H264 or HEVC so far released indicate the iMac Pro has hardware acceleration from somewhere. Besides QS this could be the fixed function logic on the Vega card itself since AMD has specialized logic unrelated to the GPU called Unified Video Decoder (UVD) and the encoder is VCE (Video Coding Engine). But UVD and VCE to my knowledge have not been widely used by developers.

It's odd that it's so much faster than the 2017 iMac on Compressor export of ProRes to HEVC since that presumably uses Quick Sync -- but we don't know for sure. It's possible they picked a Compressor HEVC export preset which is can't use Quick Sync, such as multi-pass encoding. Or maybe the Vega64 fixed-function HEVC encode hardware is just that much faster. Eventually these things will be known.

In the interim it's good news the iMac Pro either has Quick Sync or it's getting hardware encode/decode acceleration from another source (at least for FCPX and Compressor). It's very fast either way, and doesn't apparently suffer the typical Xeon encode/decode performance problem of the nMP.

 

[cynics]

I think for the sake of Apple getting similar resulting file quality they would have used multi-pass.

This would have benefited them with that benchmark but I don't think its unfair either. Regardless that compressor benchmark is otherwise pointless without very specific details that even the fine print doesn't explain.

 

[EugW]

As I mentioned a few weeks ago in another thread when it was reported that Apple would have an ARM chip in the iMac Pro, hardware HEVC acceleration could also be coming from that Apple A10 Fusion chip.

That would save them from having to implement a third form of hardware acceleration. They already use Quick Sync on Intel and their own on ARM.

I have no idea about the comparative speeds and quality though.

EDIT:

I just looked at those graphs. Using the A10 hardware accelerator makes no sense for that graph, as I wouldn't expect it to be that much faster than Intel Quick Sync, so it's probably something else. So maybe they're comparing acceleration through Vega after all then?

 

[crw]

I just looked at those graphs. Using the A10 hardware accelerator makes no sense for that graph, as I wouldn't expect it to be that much faster than Intel Quick Sync, so it's probably something else. So maybe they're comparing acceleration through Vega after all then?

The graph compares iMac Intel Media Engine (assume this means QuickSync) with iMac Pro Radeon Vega 64. That to me indicates that Compressor is accelerating via the Vega GPU. If not, why call out that it is a Vega 64.

 

[Macintosh IIcx]

It clearly states “iMac Pro Radeon Pro Vega 64” on that benchmark, so it has to be the hardware engine on the GPU.

That might only be used by Apple apps like Compressor in question or Final Cut Pro X, so don’t assume it will necessarily work in other apps unless specifically support is enabled.

Still, a very nice performance gap. I would be shocked if Final Cut Pro X isn’t massively faster than Premiere Pro on the iMac Pro.

 

[EugW]

The graph compares iMac Intel Media Engine (assume this means QuickSync) with iMac Pro Radeon Vega 64. That to me indicates that Compressor is accelerating via the Vega GPU. If not, why call out that it is a Vega 64.

Yes, I agree. I originally assumed they were just listing the overall spec but I now realize that for the CPU benches they specifically do not mention the GPU.

So yeah, Apple has implemented h.265 acceleration now on AMD... but only for the iMac Pro.

It clearly states “iMac Pro Radeon Pro Vega 64” on that benchmark, so it has to be the hardware engine on the GPU.

That might only be used by Apple apps like Compressor in question or Final Cut Pro X, so don’t assume it will necessarily work in other apps unless specifically support is enabled.

Still, a very nice performance gap. I would be shocked if Final Cut Pro X isn’t massively faster than Premiere Pro on the iMac Pro.

If they’ve implemented it for that, I’d expect QuickTime 10-bit HDR HEVC playback would also be supported. Considering I can play back those 10-bit HEVC demo videos on my 2017 Core m3 MacBook via Quick Sync in QuickTime with just 25% CPU usage, it would be a joke if an 8-core iMac Pro needed 75% CPU to play it.

Regarding QT 10-bit HEVC playback, this will serve to piss off the 2015 iMac and some of the 2016 MacBook Pro users even more. Those machines with AMD GPUs support this in hardware but Apple isn’t supporting them for this, even though now AMD based hardware h.265 acceleration is supported.

 

[Trebuin]

As I mentioned a few weeks ago in another thread when it was reported that Apple would have an ARM chip in the iMac Pro, hardware HEVC acceleration could also be coming from that Apple A10 Fusion chip.

That would save them from having to implement a third form of hardware acceleration. They already use Quick Sync on Intel and their own on ARM.

I have no idea about the comparative speeds and quality though.

EDIT:

I just looked at those graphs. Using the A10 hardware accelerator makes no sense for that graph, as I wouldn't expect it to be that much faster than Intel Quick Sync, so it's probably something else. So maybe they're comparing acceleration through Vega after all then?

Actually, you may be right. The QS technology is not built on the CPU side of the intel chips...it's built on the graphics card side of the intel chips which is rather lacking. So think about the difference between an intel graphics card & an ARM chip with video card instructions on it. Tie that in with a higher end CPU acceleration to help & maybe a third party "pro" AMD video card. I'm sure we'll find out how they pulled this off when someone tears down their iMac.

BTW, I still think non-QS encoded images are better than QS if done correctly.

 

[Macintosh IIcx]

Regarding QT 10-bit HEVC playback, this will serve to piss off the 2015 iMac and some of the 2016 MacBook Pro users even more. Those machines with AMD GPUs support this in hardware but Apple isn’t supporting them for this, even though now AMD based hardware h.265 acceleration is supported.

Arggg! I had totally forgot about that! *looks at MBP 2016 460 GPU*