Evaluation of HEVC decoders (SW, Hybrid and HW)

[rodrigo081089]

MPC? MPC-HC or MPC-BE? Install their latest version from here: MPC-HC (https://nightly.mpc-hc.org/), MPC-BE (http://mpc-be.org/builds/)
And then reset all settings (You know how to do that right?:D), ok now 4k h/w decoding is surely back. :)

Great, it was that. If i use Madvr i cant get h265 decoding by hardware.


Thank you, now i can sleep :D

[v0lt]

Nvidia GTX 950
http://i.imgur.com/pmjtdT5.png

Beauty_3840x2160_120fps_420_8bit_HEVC_MP4.mp4
decode: http://i.imgur.com/Es7Hfog.png
playback 1280x720: http://i.imgur.com/E6Kq1ko.png

It is NVIDIA CUVID mode. Tests for DXVA (native) mode here (http://forum.doom9.org/showthread.php?p=1746626#post1746626).

[wanezhiling]

@v0lt
Try UHD_PQ_Lovely_Swiss.60fps.ts which is 10bit

[NikosD]

Nvidia GTX 950

Beauty_3840x2160_120fps_420_8bit_HEVC_MP4.mp4


The first results on January by nevcairiel were slower using GTX 960.

Decode & Playback (1080p scale): 125fps

This 15% increase of GTX 950 (149-145 fps) could be drivers, better DXVA decoder or faster VP clock engine (don't think so)

[nevcairiel]

I just ran it again, and I get similar results now, 145 fps in average. Maybe they did improve drivers.
Except, I get the expected 0-2% CPU usage from a full DXVA2 decoder. Whats up with v0lt's CPU usage there?

PS:
Note that I used a debug version of the decoder, since I was too lazy to setup everything for proper benchmarking. 1-2% differences in performance are expected due to that. :)

[v0lt]

In the previous test (http://forum.doom9.org/showthread.php?p=1746607#post1746607) it seems to have been inadvertently selected NVIDIA CUVID.
Here are the results for the DXVA (native).

Nvidia GTX 950
DXVA Checker (http://i.imgur.com/qJChsHX.png)

DXVA (native) mode.

8-bit
Beauty_3840x2160_120fps_420_8bit_HEVC_MP4.mp4 (link (http://ultravideo.cs.tut.fi/video/Beauty_3840x2160_120fps_420_8bit_HEVC_MP4.mp4))
decode: 151 fps (http://i.imgur.com/TsxLyLO.png)
playback 1280x720: 151 fps (http://i.imgur.com/4JOY9Oq.png)

10-bit
UHD_PQ_Lovely_Swiss.ts (link (http://www.demo-uhd3d.com/fiche.php?cat=uhd&id=26))
decode: 121 fps (http://i.imgur.com/jcb5Bdi.png)
playback 1280x720: 121 fps (http://i.imgur.com/qJChsHX.png)

[CruNcher]

Not from the Decoder but testing the current state of the Hevc Encoder of GM204 on the GTX 970 (trying to reach highest speed)

NVEnc 1.10 (x64), using NVENC API v5.0
OS Version Windows 7 (x64)
CPU Intel Core i5-2400 @ 3.10GHz [TB: 3.20GHz] (4C/4T)
GPU GeForce GTX 970 (13 EU) @ 1266 MHz (358.78)
Input Buffers CUDA, 16 frames
Input Info y4m (yv12) -> nv12 [AVX], 1920x1080, 25/1 fps
Output Info H.265/HEVC main
1920x1080p 1:1 25.000fps (25/1fps)
Rate Control VBR
Bitrate 4000 kbps (Max: 17500 kbps)
Initial QP I:20 P:23 B:25
VBV buffer size auto
GOP length 250 frames
B frames 0 frames
Ref frames 3 frames
MV Quality Q-pel

4 Mbits 1080p (max) = 174-180 fps

[Alecex]

So the GTX 950 is just as fast for Video decoding/playback. Nice!
lets see what i II get maybe just a 950 untill new NVidia Generation

[CruNcher]

If you take it as a option into account for Media Playback it would make more sense then to go with Nvidias Android Shield TV for a bit more

[huhn]

and you are totally doomed if you want to play a file that can't be hardware accelerated.

[CruNcher]

We know what the VP8 is capable of by now through all our tests pretty much same should apply for the Android Shield TV if you have other knowledge please share :)

[huhn]

you can't play 10 bit AVC and a lot of other codecs the list is really really long, VP9 doesn't work yet and maybe is going to be hybrid which will not work on a "CPU" like the shield CPU.

there is a ref frame cap on DXVA decoding don't ask me for the number...

[NikosD]

there is a ref frame cap on DXVA decoding don't ask me for the number...

The Ref frames restrictions have all gone at least since 2009.

ATI was the last one to support Ref 16 with UVD2.2, not from the beginning but with a driver update (the HW was already capable)

[CruNcher]

you can't play 10 bit AVC and a lot of other codecs the list is really really long, VP9 doesn't work yet and maybe is going to be hybrid which will not work on a "CPU" like the shield CPU.

there is a ref frame cap on DXVA decoding don't ask me for the number...

Somehow i doubt this as it would mean no Netflix 4K support ?

[sneaker_ger]

Netflix uses HEVC for UltraHD, not 10 bit AVC or VP9.

[huhn]

netflix is HEVC isn't it?

[CruNcher]

Netflix uses HEVC for UltraHD, not 10 bit AVC or VP9.

Ah sorry auto replaced it with 10 bit HEVC ;)

ok so no 10 Bit AVC not really important for any content service platform though running currently in business terms, but yeah sad anyways for most non avg user content.

Most Anime Fansub viewer will get lost so for Nvidia, most probably they don't care much about that target group of users anyways in their calculation and decisions around VPx.

And in reality it could be enough resources left for Web Content VP9 with Hybdrid Decoding for the Shield Android TV so if it doesn't run already nicely in their Shield browser it's surely a matter of time VP9 services like Youtube will work.



Another thing i think i found another way of how the DXVA Native problem shows up for HEVC and the Lav Video Decoder on GM204 it shows way differently in its effect then in MPC-BEs OSD spikes only but it causes real visual issues in DXVA Checkers playback that look like Frame Ordering issues when they appear at certain spots at playback, interesting :)

http://i2.sendpic.org/t/8L/8LIKFDTBjHfBRArxmAZ3drMZSN9.jpg (http://sendpic.org/view/2/i/5ApYYLED2B5FvFFfmlk2IyHXi4m.png)
http://i2.sendpic.org/t/Ad/Ad3MdjNdzFxHpkCFdhCCQa6lQDB.jpg (http://sendpic.org/view/2/i/bWJqH3qK8jvyhwNusl972nhoIs.png)

As you can see it looks pretty stable from the Data but at playback it wasn't it showed Frame Ordering issues (jumpy frames forward/backward) at several places.

here in MPC-BE you wont see these Frame Ordering issues only the spikes DXVA Native is causing @ playback with Lav Video Decoder and recognize some small slowdown in this test case.

http://sendpic.org/view/2/i/o5NmFsOU8aEiG714hE5ZlgTJcR4.png




So something is different with DXVA Checker :)

DXVA Native = Frame Jumps
DXVA Copyback = No Frame Jumps
NVCuvid = Frame Jumps
None (OpenHevc) = No Frame Jumps

that is really strange that Cuvid also shows Frame Jumps and only DXVA Copy Back and Software show none.

[rodrigo081089]

The Man In the High Castle 3840x2160 H264 10bit doesnt work with DXVA

[sneaker_ger]

H.264(=AVC) 10 bit hardware/DXVA decoders do not exist in the consumer world.

[NikosD]

Impressive, but it is for 2H of 2016.

I think all big 3 (AMD, Intel, Nvidia) will have those features in 2016 - I'm not sure about VP9 10bit. I haven't seen any clip.

[wanezhiling]

AMD, I don't think so.

[Yups]

http://i.imgur.com/iPxdPTl.jpg

Kaby Lake supports fixed function HEVC 10bit and VP9 10bit decoding.



It's a Gen 9.5 graphics by the way. It might support HDMI 2.0 as well.

[huhn]

AMD, I don't think so.

i'm pretty sure they will support it. there is no reason to skip it and they don't relabel again...

in the past they where the first with bitstreaming and 5.1 PCM.

[Alecex]

VP9 would be good for chrome youtube but they might set it "on" with the drivers at some point...

I now have a palit gtx960 4GB Jetstream because its so cool and quiet.
But There is a Problem. When I play 4k Youtube Videos in Mozilla Firefox a get micro stutter. My CPU stays at very low usage while the 960 is getting the load.
Thats the video
https://www.youtube.com/watch?v=TmDKbUrSYxQ
It happens every 1 second. It seems to be like the PullDown Problem we had in the early days of 24p BluRay when wie didnt have tru 24p TVs

[CruNcher]

No issues here on Firefox 45 internal Player for the H.264 stream with the HTML5 player but the quality of the Moving Watter Surface encoding for a 4K result is really awefull :(
You really would want to avoid something like this for 4K

[Alecex]

sure its awefull but at least it does play hardware accelerated and my computer does not run hot... my gtx 460 did not accelerate anything above 1080p and it also got loud with any acelerated video...

how do i see if its the html 5 player or the flash? edit... foudn it

[Yups]

i'm pretty sure they will support it. there is no reason to skip it and they don't relabel again...

in the past they where the first with bitstreaming and 5.1 PCM.


You can't be sure without any reliable leak. As for APUs first chance it will happen is sometime in 2017 with a Zen based APU.

[daWsOn_s]

Was anyone able to get a real time playback on i5 intel ivy bridge for 4K HEVC?

I'm using the i5-3570k overclocked at 4Ghz and with mpc-hc and lav I can't get a decent playback.

Thanks.

[Alecex]

Do the Skylake i5 6200U and so on with the HD Graphics 520 have real HW VP9 h.265 Decoder like the Desktop Skylakes? If so has any one benched them?

[P.J]

Was anyone able to get a real time playback on i5 intel ivy bridge for 4K HEVC?

I'm using the i5-3570k overclocked at 4Ghz and with mpc-hc and lav I can't get a decent playback.

Thanks.

Which video exactly?

[NikosD]

New beta drivers from Intel v4326 - they cover even Win 10 Nov Update (10586)
https://downloadcenter.intel.com/download/25541/Intel-Beta-Graphics-Driver-for-Windows-7-8-1-10-15-40-

igfxHK module crashed during installation and after restart too - I disabled "hotkeys" from the Control Panel.

The drivers support MediaSDK API v1.17 and probably VP8_VLD.

Can someone using latest Win 10 Nov Update with Broadwell or Skylake report with DXVA Checker if VP8_VLD is present and if there are any new VP8/VP9 decoders (using Check Decoders -> Select format from DXVA Checker) ?

[CruNcher]

GPU GeForce GTX 970 (13 EU) @ 1266 MHz (359.00) (GM204) WDDM 1.1

H.265 Encoder PCIE 2.0 Transcoding (CPU Decoding/GPU Encoding)

4 Mbits 1080p (max 17.5 Mbits) = 181 FPS

Not sure if anything changed on the Hybdrid Decoding side doesn't look like it though from the First tests 4K Decoding Performance seems still slightly behind 8 bit OpenHevc 64Bit on 4 CPU Cores (Sandy Bridge TB: 3.20GHz no HT) :(

I guess Nvidia stopped Optimizing the Maxwell Shader Decoder so most probably we wont see any more gains here for VP6, i at least hoped they would reach similiar OpenHEVC Performance then Intel 4 Core Sandy Bridge IPC alone but they don't reach it for the same Extreme Bitstream complexity Levels at least not on the GTX 970 13 EUs.

Maybe though they reach it under Windows 8.1 (WDDM 1.3) or 10 (WDDM 2.0) can't say that yet :)

[wanezhiling]

http://xhevc.com/en/downloads/downloadCenter.jsp
6.The 10-bit encoded HEVC video can be correctly decoded (but still rendered in 8-bit for output).
Now on my laptop 4k hevc 10-bit 60fps can be played smoothly, with FFmpeg its just horrible.

[CruNcher]

WOW what for a CPU ? also Atemes Broadcast Streams complexity (Samsung Korea) ?

[sneaker_ger]

I don't see much of a difference between Lentoid 32/64 and LAV 64. As usual, only LAV 32 pales in comparison. Mainly interesting for those stuck to 32 bit software like DVB Viewer.

Tested on Core i7-860 (SSE4.2, no AVX), GraphStudioNext, Null Renderer:

UHD_PQ_Lovely_Swiss.ts (10 bit)
lav 64: 47.7 fps
lentoid 64: 45.1 fps
lentoid 32: 43.8 fps
lav 32: 19.0 fps

Ducks.Take.Off.2160p.h.265-Zion.mkv (8 bit)
lav 64: 39.9 fps
lentoid 64: 39.3 fps
lentoid 32: 38.0 fps
lav 32: 20.1 fps

Sad to see LAV's speed hasn't really improved in over a year, at least not for non-AVX CPUs. Seems like the OpenHEVC optimizations still haven't been ported to ffmpeg?

[P.J]

http://xhevc.com/en/downloads/downloadCenter.jsp

Now on my laptop 4k hevc 10-bit 60fps can be played smoothly, with FFmpeg its just horrible.

It doesn't even support DXVA2 :confused:

[NikosD]

New updated CPU results for LAV, Lentoid and Microsoft's H.265 decoder

System:
Win 10 x64 (10586) - Core i7-4790 - iGPU HD4600@1.5GHz - v.4326

Decoders:
LAV x64/x86 0.67.3
Lentoid x64/x86 2.1.0.0
Microsoft HEVC decoder build 10586

Tools:
DXVA Checker 3.10.0 in Benchmark Decode mode

Splitter:
LAV Splitter x64/x86 0.67.3



1.ProRes-1080p@30fps-2Mbps



Lentoid x64 320/398/460 CPU 84%

LAV x64 309/396/467 CPU 78%

Lentoid x86 280/380/453 CPU 84%

MS HEVC x64 280/362/410 CPU 87%

MS HEVC x86 165/222/275 CPU 51%

LAV x86 126/166/227 CPU 87%




2.Elephants Dream-1080p@24fps-1.7Mbps



MS HEVC x64 139/360/490 CPU 85%

Lentoid x64 120/352/594 CPU 63%

LAV x64 123/328/565 CPU 55%

Lentoid x86 113/302/490 CPU 65%

MS HEVC x86 77/260/415 CPU 52%

LAV x86 66/145/360 CPU 59%



3.Big Buck Bunny-1080p@60fps-2Mbps



MS HEVC x64 280/410/490 CPU 86%

LAV x64 231/400/560 CPU 52%

Lentoid x64 169/386/530 CPU 57%

Lentoid x86 193/340/490 CPU 61%

MS HEVC x86 178/333/442 CPU 53%

LAV x86 71/162/365 CPU 53%



4. Girls -1080p60 fps -11Mbps


MS HEVC x64 223/290/348 CPU 91%

Lentoid x64 199/284/355 CPU 81%

LAV x64 187/269/335 CPU 78%

Lentoid x86 190/260/330 CPU 82%

MS HEVC x86 126/191/252 CPU 52%

LAV x86 94/133/190 CPU 81%


5.Beauty-2160p@30fps-12.3Mbps



Lentoid x64 69/85/91 CPU 67%

LAV x64 71/84/89 CPU 69%

MS HEVC x64 71/80/83 CPU 77%

Lentoid x86 63/79/86 CPU 75%

MS HEVC x86 52/59/63 CPU 47%

LAV x86 32/44/48 CPU 88%



6.Fitness-2160p@30fps-8Mbps



LAV x64 93/107/128 CPU 81%

Lentoid x64 66/99/127 CPU 89%

Lentoid x86 66/90/118 CPU 90%

MS HEVC x64 70/89/110 CPU 88%

MS HEVC x86 52/67/92 CPU 51%

LAV x86 37/58/105 CPU 91%



7.Ducks-2160p@50fps-4Mbps



LAV x64 103/113/116 CPU 86%

Lentoid x64 85/100/105 CPU 78%

MS HEVC x64 91/98/100 CPU 90%

Lentoid x86 73/89/92 CPU 89%

MS HEVC x86 73/78/81 CPU 55%

LAV x86 37/46/48 CPU 91%


8.Gravity - 2048x858@24fps-5Mbps-10bit



Lentoid x64 127/196/353 CPU 76%

LAV x64 119/187/393 CPU 60%

Lentoid x86 118/179/354 CPU 79%

MS HEVC x64 99/150/307 CPU 93%

LAV x86 74/120/348 CPU 66%

MS HEVC x86 58/92/308 CPU 51%


Comments:


1) Lentoid added 10 bit decoding with 8bit output (I420).
MS HEVC decoders 10bit decoding uses 8bit output too (NV12)

2) Lentoid x86 keeps running faster than LAV x86 by a huge margin and it's faster than MS HEVC x86 too.

3) Lentoid x64 is very close to LAV x64 and generally faster but there are strong fluctuations in the results, so sometimes it's difficult to measure both decoders.

4) The real surprise is MS HEVC x64 which is faster than anything, but only for 1080p.
In the first clip, it has some problems, it even crashes DXVA Checker when GPU acceleration is used.

But for all the other 1080p clips is definitely faster.
On the other hand, MS HEVC decoders are slow on 2160p clips and especially 10bit clips

It seems - most of the times - that whenever each decoder manages to utilize more CPU usage, it ends first in the specific clip.

[Manni]

Hi everyone,

At the moment I have an AMD HD7870 which struggles with many UHD/4K files, especially 10bits ones.

I had ruled out upgrading to the AMD Fury/Fury X due to the lack of HDMI 2.0 (18Gb/s) and HDCP 2.2 support, but this looks like it's going to be solved by the upcoming Club3D displayport 1.2 to HDMI 2.0 active adapter (http://club-3d.com/index.php/products/reader.en/product/displayport-12-to-hdmi-20-uhd-active-adapter.html). I could wait for Arctic Islands, but waiting until at least Q2 2016 isn't a great prospect.

So my question is: would a Fury (I don't think I'll stretch to the Fury X as this is mainly for better video with MPC-BE/LAV/MAdVR) significantly improve this?

Reading the thread I'm not sure if the Fury supports hardware acceleration for x265 (and VP9, although that's not that much of an issue for me right now).

Does anyone know what the status is with the new Crimson 15.11 driver, which should support the Club3D adapter?

My CPU (see sig) is not the latest and greatest, so I need as much GPU acceleration I can get.

I tried to find answers but I didn't find any solid info about the state of h/w or hybrid decoding for the new AMD GPUs.

Thanks!

[chros]

http://xhevc.com/en/downloads/downloadCenter.jsp

Now on my laptop 4k hevc 10-bit 60fps can be played smoothly, with FFmpeg its just horrible.
And what about LAV video x64 if only enable the 4:2:0 output? (I can't play back 10bit with 10bit output on my system (too slow for that), but I can in 8 bit.)

[sneaker_ger]

4:2:0 is chroma sub-sampling, not bitdepth. There is both 4:2:0 8 bit and 4:2:0 10 bit.

For LAV any kind of chroma resampling or 10 bit -> 8 bit dithering means additional computation, not faster processing. Either LAV or the renderer has to do this (in most cases). If your system struggles with the 10 bit output it might have to do with your renderer.

[nevcairiel]

Having LAV convert 10 to 8-bit is nearly free from a performance standpoint, so using that instead of 10-bit could result in an overall performance increase.

[NikosD]

I updated my post with MS HEVC decoder results.

What a surprise!

MS managed to reach the top once again but only for 1080p clips

I clearly remember talking about the subject, saying that the development of HEVC decoder by Microsoft hasn't ended yet, even if official Win 10 build 10240 had released.

A lot of people disagreed with me, telling me that the official Win 10 release meant the end of HEVC decoder development.

Well, we can now see I think, that HEVC decoding development still goes on for Microsoft reaching the top once again.

Unfortunately, 2160p HEVC decoding is slow using MS HEVC decoders and especially in 10bit clips.

MS should put more effort on that.

[NikosD]

At the moment I have an AMD HD7870 which struggles with many UHD/4K files, especially 10bits ones.


So my question is: would a Fury (I don't think I'll stretch to the Fury X as this is mainly for better video with MPC-BE/LAV/MAdVR) significantly improve this?

Reading the thread I'm not sure if the Fury supports hardware acceleration for x265 (and VP9, although that's not that much of an issue for me right now).

Does anyone know what the status is with the new Crimson 15.11 driver, which should support the Club3D adapter?



Decoding UHD/4K H.264 & H.265 would be no problem with a Fury card as long as those clips are 8bit.
Because Fury has real HW decoder for 8bit H.264 & H.265 codecs.

4K 10bit HEVC clips could be a problem if the bandwidth is high, like the upcoming BluRay 4K.

Regarding VP9, I don't think Fury has hybrid support, but AMD could add it to the driver in the future.

I don't know if Crimson has changed that.

Your CPU (Core i7 3770K) is fast enough for any low bandwidth clip (H.265 and VP9)

[Manni]

Decoding UHD/4K H.264 & H.265 would be no problem with a Fury card as long as those clips are 8bit.
Because Fury has real HW decoder for 8bit H.264 & H.265 codecs.

4K 10bit HEVC clips could be a problem if the bandwidth is high, like the upcoming BluRay 4K.

Regarding VP9, I don't think Fury has hybrid support, but AMD could add it to the driver in the future.

I don't know if Crimson has changed that.

Your CPU (Core i7 3770K) is fast enough for any low bandwidth clip (H.265 and VP9)

Thanks a lot for this.

The demo clips I'd like to be able to test are all under 50mb/s. They are detected as BT2020 by MadVR but it's just the container, they are most likely rec709 or P3.

UHD Bluray, we'll see, it causes other issues like P3 gamut, HDR, etc. Plus it might take a while before the new AACS is cracked or PDVD supports it. I'll probably use a standalone player for a while anyway just for these.

I guess if a Fury helps me to upscale x264 bluray to UHD with full MadVR quality in 10bits 4:4:4 and to play a few HEVC demos I'll be happy until I can get an Arctic Island towards the end of 2016.

[envagyok]

At the first post, what mean the three numbers end of lines?

1.ProRes-1080p@30fps-2Mbps
https://www.sendspace.com/file/mo8exg


LAV x64 i7-4790 158/230/251

LAV x86 i7-4790 121/151/188

LAV x64 i7-4790 (DXVA) 83/134/157

LAV x86 i7-4790 (DXVA) 84/133/158

Strongene x86 i7-4790 (OpenCL) 83/112/146

Strongene x86 i7-4790 97/109/111

LAV x64 i5-4200M 76/98/114

LAV x64 i5-4200M (DXVA) 60/84/96

LAV x64 Core 2 Quad 45/52/54

[NikosD]

The three numbers always mean:

Minimum/average/maximum number of frames per second (fps)

The more the the faster, obviously.

[Nullack]

In trying to get an idea of UHD bluray decoding for when discs are purchasable, I've so far found a sample with HEVC Main 10 L52 2160P60 using a constant bit rate of 51.5 Mbps. Using my GTX 960 with nightly MPC-HC x64, nightly Lav Filters x64, in DXVA 2 native mode on the EVR-CP renderer, all is good with only minor paint times maximums around 25ms against the 60 fps cadence of 16 ms. The great majority of frames are being done in around 5-6 ms, which is heaps of headroom below the 16 ms 60 fps cadence. The statistics aside to the eye it all looks nice and smooth. Is there any other sample at a higher bit rate that would be closer to test what the UHD bluray discs might be like?

[NikosD]

Is there any other sample at a higher bit rate that would be closer to test what the UHD bluray discs might be like?

This is a 10bit HEVC 2160p50 fps test clip at 127Mbps:
https://www.sendspace.com/file/ietm9n

This a HDR 10bit HEVC 2160p60 fps clip at 45Mbps:
http://demo-uhd3d.com/files/uhd4k/NEO_Fire_HDR_HEVC_2160p_5994_Main10_45-100VBR.mp4

Give them a try and tell us your experience.

[chros]

Is there any other sample at a higher bit rate that would be closer to test what the UHD bluray discs might be like?
But won't movies/tv shows be (stay) 24p?

[Nullack]

This is a 10bit HEVC 2160p50 fps test clip at 127Mbps:
https://www.sendspace.com/file/ietm9n

This a HDR 10bit HEVC 2160p60 fps clip at 45Mbps:
http://demo-uhd3d.com/files/uhd4k/NEO_Fire_HDR_HEVC_2160p_5994_Main10_45-100VBR.mp4

Give them a try and tell us your experience.

Thanks. That short clip of the runners, under the same config I described previously that 127 Mbps bitrate makes no difference in the telemetry as my 60 FPS L52 MAIN 10 sample. Meaning, the same video engine utilisation, the same GPU core utilisation, the same gpu memory usage. Also the EVR-CP jitter and paint time stats are around the same too. So basically that setup gobbles up the video of the runners race no sweat.

The BT.2020 took more fiddling to test because I had to switch my colour mode in my UHD 4K display and that required removing the HDMI, changing the setting, putting it back then changing the desktop settings too. Once all the colour was configured OK. I had to drop down from 60 hz to be able to do the BT.2020 HDR over HDMI 2 - my display only has HDMI for UHD. I played it back under the same config and it made no difference to the EVR stats or the telemetry of how the GPU was loaded. So it seems that with HEVC 4K playback, the HDR BT.2020 basically comes for free in terms of measuring its impact on the GPU and the renderer stats in comparison to "normal" L52 MAIN 10 HEVC.