HEVC 10 bit export

[aagribeiro]

Hi,

I'm testing a workflow to export to HEVC 10 bit.

I got some 4K 10 bit sample files from a Sony A7S III, did a small FCPX project and did File->Share to a 10 bit HEVC compressor settings destination. A 30 second clip took 10 minutes to export on my Mac Pro (16 core, W5700X). I was expecting this, since I'd read that hardware encoding wasn't supported in FCPX.

I then exported this project to ProRes 422 HQ, opened it in QuickTime, and exported to HEVC. It exported a 10 bit HEVC file in around 38 seconds.

Is QuickTime exporting using the hardware encoder of the W5700X? Does this mean that we soon may have 10 bit HEVC hardware encoding on Compressor?

Thanks,
Alex

 

[fhturner]

Are you sure the QuickTime export was 10-bit? Whenever I use QT Player, there's only an HEVC checkbox, no other settings available. So I'm wondering if it used 8-bit, which was hardware-accelerated, while 10-bit in Compressor, for some reason, was not.

 

[aagribeiro]

Yes, I'm sure, that was my first thought. I have a ProDisplay XDR, and the difference between 8bit and 10bit is night and day. Additionally I copied that file to my Plex server and saw it on an LG Oled, and it was HDR. Finally, I exported an HEVC 8bit file in FCPX (got a warning because I was exporting HDR in an 8 bit codec), and the result was awful.

 

[Ritsuka]

Hardware encoders are set in stone (or better, in silicon). Your CPU/GPU has got a number of fixed hardware encoders for a number of formats.

Your CPU/GPU hasn't got a 10bit HEVC encoder. The only way to change that is to buy a new CPU or GPU that has got one (if there is one on the market).

ProRes is a simpler format, it doesn't need much cpu time to be encoded/decoded.

 

[aagribeiro]

If I use the Pro Res file in an FCPX project and convert it to 10 bit H.265, it takes 10m, and I see no usage of the GPU in iStats.

If I use the Pro Res file in QuickTime and convert it to HEVC (which results in a 10 bit H.265 file), it takes 38 seconds and I see around 20% GPU usage in iStats.

My conclusion is that either there's a 10 bit HEVC hardware encoder on the W5700X or Metal is being used to speed up the process in QuickTime but not in FCPX.

 

[joema2]

...My conclusion is that either there's a 10 bit HEVC hardware encoder on the W5700X or Metal is being used to speed up the process in QuickTime but not in FCPX.

You are correct. There are several separate issues (1) Does the HEVC hardware encoder exist (2) Does it work for the specific encoding format? IOW some might work for 8-bit but not 10-bit. OR some might work for 4:2:0 but not 4:2:2.
(3) Does the application software use the hardware encoder?

A good example of this is comparing the 10-bit HEVC export performance of latest version of DaVinci Resolve Studio, ver 16.2.5 vs FCPX 10.4.8 on a 2019 top-spec MacBook Pro 16, also on a 2017 10-core Vega64 iMac Pro. Both machines were on Catalina 10.15.6.

Below are the numbers I got when exporting a 7 second 4k/29.97 10-bit 4:2:2 ProRes 422 file as HEVC. Obviously the hardware acceleration exists and is usable, whether that's T2, QuickSync (on MBP) or AMD's VCE. However the current version of FCPX is not using it for 10-bit export.

*** 8-bit HEVC export ***
2019 8-core MacBook Pro 16, 5500M, FCPX 10.4.8: 7.2 seconds

*** 10-bit HEVC exports ***
2019 8-core MacBook Pro, 5500M, FCPX 10.4.8: 5 min 2 sec
2019 8-core MacBook Pro, 5500M, Resolve 16.2.5: 5.6 seconds

2017 10-core Vega64 iMac Pro, FCPX 10.4.8: 2 min 50 sec
2017 10-core Vega64 iMac Pro, Resolve 16.2.5: 5.3 seconds

 

[kohlson]

Max Yurev's recent YT vid's discuss this. Apparently 10-gen Intel have this in hardware. Beats out his 2019 MacPro for 10-bit..

 

[1229175]

8-bit HEVC exports are hardware-accelerated by the T2 chip in Final Cut Pro X and Compressor. 10-bit HEVC exports are not hardware-accelerated. The encoder is software-based and only runs on the CPU.

 

[kohlson]

10-bit HEVC exports are not hardware-accelerated.

Lots of info to parse. Max Tech Review. The data sheet seems to says things about 10-bit HEVC, depending on the actual CPU used.

 

[1229175]

Lots of info to parse. Max Tech Review. The data sheet seems to says things about 10-bit HEVC, depending on the actual CPU used.

I was referring to 10-bit HEVC exports in macOS. macOS does not support the Intel-CPU- or AMD-GPU-based HEVC encoder. You'd only benefit from those in Linux or Windows.

 

[kohlson]

macOS does not support the Intel-CPU- or AMD-GPU-based HEVC encoder.

I'm not doing h265 as yet, so no personal experience on my cMP. But there seems to be others who have enabled hardware h264 and h265 encode/decode, without T2. For example this and this, which enable hardware in an appropriate GPU.

There was a reference to taking advantage of the Intel 10th-gen CPU for good video performance in Canon R5 clips in this video.

I understand that T2 will be very effective at hardware decoding/encoding. And I look forward to through reviews on this. But I would be very surprised if Apple/Macos did not take advantage of all available capabilities (CPU/GPU/T2) in their video and other graphics workloads.

 

[1229175]

I did some testing in HandBrake and both the H.264 and HEVC VideoToolbox encoders are T2-based; not GPU-based. Both are also 8-bit codecs. There's no 10-bit VideoToolbox encoder for either codec.

 

[kohlson]

While I don't have any direct experience on h265, I do have h264 8-bit hardware accelerated encoding and decoding in my cMP with RX580. It's quite noticeable in FCPX (and HB) when it becomes disabled.

 

[1229175]

While I don't have any direct experience on h265, I do have h264 8-bit hardware accelerated encoding and decoding in my cMP with RX580. It's quite noticeable in FCPX (and HB) when it becomes disabled.

Does your Mac have a T2 chip? I'm wondering now if perhaps macOS disables GPU-based encoding in favor of T2-based encoding on Macs that have it.

 

[1229175]

I just ran the same H.264 VideoToolbox test in HandBrake on a 2012 MacBook Pro without a T2 chip and it didn't use the GPU either. That's odd because I recall seeing GPU usage increase during a similar encode several months ago.

 

[kohlson]

Does your Mac have a T2 chip?

It's a 2009 Mac Mac Pro, flashed with 5,1 (2010) firmware. Designed and produced long before T1 was even a twinkle.

Macos and their apps like FCPXiGPU do a pretty good job of leveraging available resources. For example, on my late-2013 15-MBP with integrated (Iris Graphics) and dGPU (Nvidia 750M) you can see both working on import/timeline and especially export.

And when Intel started including QuickSync, Apple leveraged that, too. But the challenge came with iMac Pro, which uses Xeon (no QuickSync). So they wrote some code to enable using the GPU hardware to handle encode/decode. Some delightful cMP owners figured out was was done, and have shared this knowledge with this community.

QuickSync is not comprehensive, and doesn't handle all formats and encodings. An example of this is some of the newer 10-bit HDR Canon codecs. But Intel has updated some of this in their 10th-gen CPU. However, and I am still learning about this, it appears that Intel is moving too slowly in this regard for Apple's taste, and T2 (or T whatever) is the future. But it is hard to imagine that AMD won't be part of the answer, too.

 

[joema2]

I was referring to 10-bit HEVC exports in macOS. macOS does not support the Intel-CPU- or AMD-GPU-based HEVC encoder. You'd only benefit from those in Linux or Windows.

Actually, MacOS Catalina supports this but it's up to the app to take advantage of it. A good example is the latest version of DaVinci Resolve Studio 16.2.5 on both iMac Pro and 2019 MacBook Pro 16 is very fast at 4k 10-bit HEVC export. Ironically FCPX 10.4.8 is slow; for some reason it doesn't yet use harware encode acceleration for that case.

Whether it's using Quick Sync, T2 or AMD's VCE, I don't know. But it's obviously hardware accelerated because it's about 30x faster.

Earlier versions of Resolve (within the past six months) were roughly as slow as FCPX, so this is a recent change. The good news is this indicates it's not a fundamental MacOS or hardware issue but FCPX needs updating.

Premiere 14.3.2 on the same MacOS and hardware says it doesn't support 10-bit HEVC encode acceleration but it's quite fast. It may be using h/w accel and just not reporting it correctly in the UI.

 

[1229175]

Actually, MacOS Catalina supports this but it's up to the app to take advantage of it. A good example is the latest version of DaVinci Resolve Studio 16.2.5 on both iMac Pro and 2019 MacBook Pro 16 is very fast at 4k 10-bit HEVC export. Ironically FCPX 10.4.8 is slow; for some reason it doesn't yet use harware encode acceleration for that case.

Whether it's using Quick Sync, T2 or AMD's VCE, I don't know. But it's obviously hardware accelerated because it's about 30x faster.

Earlier versions of Resolve (within the past six months) were roughly as slow as FCPX, so this is a recent change. The good news is this indicates it's not a fundamental MacOS or hardware issue but FCPX needs updating.

Premiere 14.3.2 on the same MacOS and hardware says it doesn't support 10-bit HEVC encode acceleration but it's quite fast. It may be using h/w accel and just not reporting it correctly in the UI.

You're right. I did some testing and it certainly is comically ironic that Final Cut Pro X and Compressor don't support using the hardware encoders for 10-bit HEVC while DaVinci Resolve does. That leads me to concluded that it's not macOS, but macOS's default encoder (formerly part of QuickTime and now VideoToolbox) that doesn't support the hardware-based encoding for 10-bit HEVC.

 

[SWAON]

Can this be done on the new M1 chip devices for testing?

 

[aagribeiro]

I was just testing this with HDR footage from an iPhone and in Compressor, if the encoder setting is set to Fast, the GPU is used (at least the W5700X I have) and the encoding is much faster.