Evaluation of HEVC decoders (SW, Hybrid and HW)

[Yups]

VP8 shows exactly the same.

[NikosD]

Weird...It should say VP8 according to the previous version of DXVA Checker and according to the installed decoders.
Unless of course it is somehow common.

[P.J]

Using latest DXVA Checker v3.70 and your HD 530, if you go to "Check Decoders" -> "Select Format" and you choose VP8 or VP9, are there any other decoders besides LAV ?

I did it for HEVC but got no HEVC_VLD_Main10 for LAV:

http://i.imgur.com/fEslJSK.png

[sneaker_ger]

Does it only show native DXVA? Then you might need a recent nightly version of LAV.
http://files.1f0.de/lavf/nightly/

[P.J]

Does it only show native DXVA? Then you might need a recent nightly version of LAV.
http://files.1f0.de/lavf/nightly/

Still no luck.

[NikosD]

No decoder listed. I don't have VP8/9 samples.




Here it is a VP8 sample:
https://www.sendspace.com/file/be21aw

And a VP9 from Youtube:
https://www.sendspace.com/file/v95vjb

[oscarbg]

After a bios update:


decode= 159-184-212 3%
playback= 148-170-203 3-4%

can you please post full system specs (motherboard model, bios versions,etc..)

[oscarbg]

Guys isn't time to test HW accelerated HEVC encode?
It is present on Skylake and NV Maxwell Gen 2 GPUs (yes even my GTX 970 which doesn't support HEVC decode)..
One way to test will be ffmpeg which supports using NVENC for Nvidia support and QuickSync for Intel..
We should make standard binary and command line args with a standard video input too similar to h264 benchmark (Tech ARP - x264 HD Benchmark Ver. 5.0.1)..
Then let the benchmarks begin..

[NikosD]

I have been thinking of this concept since last August when I purchased my Core i7 4790.

But most people are interested in SW encoding and particularly SW quality encoding, not speed.

So, a very specific benchmark test regarding speed of HW encoding and expecially HEVC, which is so immature right now, it doesn't make a lot of sense to me.

But you are free to open such a thread, I don't think there is something like that in this forum.

I know only a quality comparison thread between x264 and x265.

[P.J]

It's kinda useless since SW encoding provides better quality and nobody wants to encode in real-time on PC.

[huhn]

It's kinda useless since SW encoding provides better quality and nobody wants to encode in real-time on PC.

real time encoding is very important for live streaming such as streaming computer games to TVs and things like that.

if HEVC hardware encoding is better than the old AVC hardware encoder they would be really useful for such things.

the pictures i saw from the nvenc HEVC encoder wasn't good at all.

[NikosD]

I did it for HEVC but got no HEVC_VLD_Main10 for LAV:



Talking with DXVA Checker's developer, he told me that the new 3.7.2 version has fixed that (not released yet)

No decoder listed. I don't have VP8/9 samples.


The VP9and8 DXVA device decoder has been removed because it's not supported by Intel (yet or ever!)

Also the developer of DXVA Checker told me, that it's impossible to use DXVA Checker for VP9/VP8 samples with Intel MFT decoder, because it simply is not working (yet or ever!)

[NikosD]

For anyone interested in testing, I encoded a 8K (7680x4320) HEVC clip at 60fps in both 10bit and 8bit versions.

Of course, you can't HW accelerate the decoding of those clips even if you have a 960 card, due to high resolution.

You need a very fast CPU.

The clips are just 1 sec and very small.

Clip 1- HEVC 4320p60fps -10bit


Clip 2 - HEVC 4320p60fps -8bit





@Yups

Back on February of 2015, a few months ago, I have posted two 8K (!) HEVC clips and a lot of people could think that I'm crazy to post such high-resolution HEVC clips, since not even 4K is largely populated yet.

BUT as I read in Intel forums, Skylake supports in HW the 8K resolution for HEVC (!!) (not for H264)

I uploaded again those two clips in 10 bit (that you can't test in pure HW) and a 8bit HEVC 8K clip that I would like you to test.


The clips are just 1 sec and very small.


Clip 1- HEVC 4320p60fps -10bit
https://www.sendspace.com/file/wglrw4

Clip 2 - HEVC 4320p60fps -8bit
https://www.sendspace.com/file/difznd


I would like to see some benchmarks in both modes (Decode and Playback) with DXVA native using DXVA Checker.

Also, it would be interesting to test Skylake CPU and hybrid GPU with 8K HEVC 10bit clip.

Thanks!

[Yups]

This one should be better for Yups to test: http://demo-uhd3d.com/fiche.php?cat=uhd&id=97

Hybrid has terrible hiccups though.

Hybrid

Decode= 56-74-144 [22-36%] ~110 watts
Playback= 48-62-106 [0-28%] ~107 watts


CPU only

Decode= 83-99-156 [68-98%] ~130 watts
Playback= 62-70-81 [35-97%] ~127 watts


i7-6700k 4 Ghz @1.15V
HD 530 @1150 Mhz 10.18.15.4274


CPU only is faster in this case, but all 8 threads are running with ~90% load most of the time. I think on an ULT GT2 SKU with 4 threads only Hybrid will be faster than pure CPU.



Because someone asked, here 4 CPU cores without HT.


Hybrid

Decode= 56-76-144 [37-78%, avg 51%] ~102 watts
Playback= 51-62-108 [28-54%, avg 41%] ~97 watts


CPU only

Decode= 63-72-99 [72-99%, avg 92%] ~105 watts
Playback= 58-70-98 [73-98%, avg 90%] ~105 watts

[NikosD]

Did you see my post above with 8K HEVC (10bit/8bit) clips ?

[Yups]

@Yups

Back on February of 2015, a few months ago, I have posted two 8K (!) HEVC clips and a lot of people could think that I'm crazy to post such high-resolution HEVC clips, since not even 4K is largely populated yet.

BUT as I read in Intel forums, Skylake supports in HW the 8K resolution for HEVC (!!) (not for H264)

I uploaded again those two clips in 10 bit (that you can't test in pure HW) and a 8bit HEVC 8K clip that I would like you to test.


The clips are just 1 sec and very small.


Clip 1- HEVC 4320p60fps -10bit
https://www.sendspace.com/file/wglrw4

Clip 2 - HEVC 4320p60fps -8bit
https://www.sendspace.com/file/difznd


I would like to see some benchmarks in both modes (Decode and Playback) with DXVA native using DXVA Checker.

Also, it would be interesting to test Skylake CPU and hybrid GPU with 8K HEVC 10bit clip.

Thanks!


Not sure how reliable a 1 second video is regarding fps but the 8 bit HEVC sample is working in hardware. 60 fps decode, 55 fps playback with low CPU load. Pure CPU decoding/playback is much slower. The 10 bit sample crashes in DXVA Checker.

[NikosD]

Thanks!

You confirm that is working and that's very good.

Does the 10bit video crash with MPC-HC or other player or DXVA checker only ?

Does it crash also with MS MFT using decode mode (the only usable HW mode) or Cyberlink H265 DXVA/OCL decoder ?

I want to confirm that it's a DXVA Checker problem and not decoders problem.

[NikosD]

Next DXVA Checker will check for 8K or 8KFUHD as it is called (Full Ultra HD)

[Yups]

Crash/hangs happens while loading/searching lav video decoder. There is no crash when I disable hardware decoder in lav settings.

[NikosD]

Crash/hangs happens while loading/searching lav video decoder. There is no crash when I disable hardware decoder in lav settings.

Can you try MS MFT H265 decoder (with "Enable DXVA" setting) in "Decode Benchmark" mode for 10bit 8K HEVC ?

Thanks!

[P.J]

Because someone asked, here 4 CPU cores without HT.


Hybrid

Decode= 56-76-144 [37-78%, avg 51%] ~102 watts
Playback= 51-62-108 [28-54%, avg 41%] ~97 watts


CPU only

Decode= 63-72-99 [72-99%, avg 92%] ~105 watts
Playback= 58-70-98 [73-98%, avg 90%] ~105 watts

:thanks:

[foxyshadis]

Completely off-topic posts about DX12 gaming moved to the PC hardware forum. At least keep this to decoding in some way, even if it's not HEVC specific anymore.

[P.J]

http://www.thedailynote.net/2014/05/why-4k-video-looks-better-on-1080p-monitors.html

[NikosD]

IIRC DXVA accelerates only 4:2:0 clips

[leonccyiu]

@Yups

Back on February of 2015, a few months ago, I have posted two 8K (!) HEVC clips and a lot of people could think that I'm crazy to post such high-resolution HEVC clips, since not even 4K is largely populated yet.

BUT as I read in Intel forums, Skylake supports in HW the 8K resolution for HEVC (!!) (not for H264)

I uploaded again those two clips in 10 bit (that you can't test in pure HW) and a 8bit HEVC 8K clip that I would like you to test.


The clips are just 1 sec and very small.


Clip 1- HEVC 4320p60fps -10bit
https://www.sendspace.com/file/wglrw4

Clip 2 - HEVC 4320p60fps -8bit
https://www.sendspace.com/file/difznd


I would like to see some benchmarks in both modes (Decode and Playback) with DXVA native using DXVA Checker.

Also, it would be interesting to test Skylake CPU and hybrid GPU with 8K HEVC 10bit clip.

Thanks!

Wow! Does the Skylake Core M with Intel HD Graphics 515 Also do 8k HEVC?

http://www.notebookcheck.net/Intel-HD-Graphics-515.149941.0.html

But since there's no 10 bit HEVC and 8k will likely have frame rates of at least 60, not to mention other codecs like vp9 which a core m isn't going to decode 4k60 in, not sure how useful it is. I heard maybe Kaby Lake and Gen10 will have 10bit HEVC decode.

[leonccyiu]

Not sure how reliable a 1 second video is regarding fps but the 8 bit HEVC sample is working in hardware. 60 fps decode, 55 fps playback with low CPU load. Pure CPU decoding/playback is much slower. The 10 bit sample crashes in DXVA Checker.

If it's 55fps playback for the 60fps videos, would that be frames being dropped? and therefore 8k60 is not possible unlike skylake?

[Yups]

Probably Kaby Lake includes 10bit HEVC but because it's in 14 nm process again, maybe they have to cut something else to fit 10 bit HEVC.


SKL 4+2 is only ~120 mm² big...plenty of die size headroom for improvements on 14nm.

[P.J]

Because someone asked, here 4 CPU cores without HT.


Hybrid

Decode= 56-76-144 [37-78%, avg 51%] ~102 watts
Playback= 51-62-108 [28-54%, avg 41%] ~97 watts


CPU only

Decode= 63-72-99 [72-99%, avg 92%] ~105 watts
Playback= 58-70-98 [73-98%, avg 90%] ~105 watts

Would you test with DDR4 2133 too? :o

[P.J]

http://www.computerbase.de/2015-09/amd-radeon-r9-nano-test/6/

[NikosD]

Tremendous processing power of GPU shaders could make you think that R9 Nano has pure HEVC HW decoder, but it doesn't.

Or more carefully I would say it shouldn't.

It's two times faster of the Nvidia 980 in 4K HEVC decoding.

[nevcairiel]

But Nano and Fury (X) do have a HEVC hardware decoder - limited to 8-bit only, but still hardware.
I dare say a hybrid decoder couldn't reach such speed. It would be CPU limited before the shaders could even all be used.

[Sulik]

I dare say a hybrid decoder couldn't reach such speed. It would be CPU limited before the shaders could even all be used.

Yeah, all hybrid decoders are fundamentally limited by CABAC on the CPU: they really only work well at low bitrates. If you go to low enough bitrates, the hybrid decoders can probably surpass pure HW decoders, but that doesn't mean much.
That's similar to software vendors showcasing their decoders -> anyone remembers the DivX 4K streams at 2Mbps ? :)

[NikosD]

I remembered commenting here the presence of HEVC decoding block inside Fury cards and the lack of HDMI 2.0 but probably because I have never seen benchmark results of HEVC for AMD I was under the idea that only Carizzo can decode in HW HEVC format.

Isn't there someone with a Fury card or a Carrizo laptop to do some HEVC tests, BTW ?

For the hybrid results I agree 100% with Sulik.

All of my tests show a full usage of shaders for low powered GPUs like HD 4600 and low bitrate clips.

For those low bitrate HEVC clips a very powerful card like R9 Nano - if it didn't have a HEVC fixed function block - could reach or even surpass a fixed function unit which can clearly show its teeth in high and very high bitrates.

[NikosD]

A new graphics department from AMD, called Radeon Technologies Group could easily be sold. That or AMD itself.

And who is gonna buy them, AMD and Radeon Group ?
Google ? Apple ? Samsung ? Or even Qualcomm ?

All I know is that I'm gonna buy after six years (!) a discrete card and it's going to be AMD or Radeon Technologies Group (or whatever they called)

I'm gonna buy AMD Greenland next-gen card on 2016, especially if it has HEVC 10bit and hopefully VP9 in HW.

http://anandtech.com/show/9612/amd-reorganizes-graphics-division-radeon-whole-once-more-led-by-raja-koduri

[baii]

Any chance to update SW and hybrid result for high bit rate clips? or they just struggle?
CPU hw decoder seem so gimped as they don't do 10bit where 960 is too weak to do needi3 ~

[NikosD]

I think it"s pointless.

CPU and hybrid decoders are not built for such tasks, like high bitrate decoding, although the iGPU of Skylake did better than I thought so on 10bit HEVC.

[aufkrawall]

Intel seems to know what they are doing, even HEVC 8 bit hybrid decoder of Haswell wasn't slow at all.
They probably get a lot of useful findings from their Quick Sync development.
Since they basically have yearly hardware updates, there's not much pressure to put everything in hardware at once.

[huhn]

so you have to buy every year?

[foxyshadis]

You buy as soon as something comes out that has improved performance enough to interest you. The dilemma that's faced hardware upgraders forever.

[huhn]

for the next years all we need is a hardware decoder that can decode 10 bit UHD 60 HZ

but adding 8 bit HEVC is like adding nothing at all.
that's why i'm not happing at all what intel is thinking.
same for AMD no HDMI 2.0 no 10 bit HEVC.

it's just terrible that you have to buy an nvidia 960 or 950.

and of cause nothing is older than the hard you just brought.

[aufkrawall]

Well, don't want to sound like a smart ass, but that's just the result of market economy with weak competition (AMD CPUs/APUs).
They are giving customers a reason to buy also next year when they probably release Kabylake (?) way before AMD Zen.

[baii]

Maybe a silly question, do any of the hw decoder run in 32bit player?

Sent from my SM-T700 using Tapatalk

[NikosD]

ALL of them.

[Yups]

Well, don't want to sound like a smart ass, but that's just the result of market economy with weak competition (AMD CPUs/APUs).
They are giving customers a reason to buy also next year when they probably release Kabylake (?) way before AMD Zen.


It's not even clear if Kabylake brings 10 bit support and HDMI 2.0. If this is just a Skylake refresh like Devil's Canyon, means Skylake CPU+Gen9 GPU, it is unlikely.

[aufkrawall]

Well, then Intel would probably be outperformed by next year's AMD / Nvidia GPU generation.
Why would they allow this to happen? I can hardly believe that they are failing technically to implement native HEVC 10 bit decoding when Haswell was so far ahead, in comparison to NV and especially AMD.

[Yups]

In regards to technically, dedicated GPUs will always have a big time advantage over a CPU with GPU, an entire platform with new chipsets/sockets and all the neccessary drivers because of a much longer development process with much longer evaluating times. There is a reason why AMDs GPU in their APUs was a generation behind their dedicated GPUs in the past. GPU only is much more flexible.

[NikosD]

DXVA checker v3.80 is out with 8K test and LAV Video 10bit HEVC support in "Select Format" option

[aufkrawall]

In regards to technically, dedicated GPUs will always have a big time advantage over a CPU with GPU, an entire platform with new chipsets/sockets and all the neccessary drivers because of a much longer development process with much longer evaluating times. There is a reason why AMDs GPU in their APUs was a generation behind their dedicated GPUs in the past. GPU only is much more flexible.
They will have had three years time to upgrade their Haswell VPU to native HEVC 10 bit decoding.
Which should be quite enough, since NV has it since January 2015.

AMD is terribly short on r&d financial resources, I wouldn't compare this to Intel.

[foxyshadis]

I know most people use 3rd party players, but saw this tidbit in a leaked Microsoft Win10 release notes for last week and figured someone might be interested:

Media: We made changes to ensure that a high efficiency video coding (HEVC) can be played in the Movies & TV app when decoder bit depth downsampling is recommended by a Direct3D version 11 user-mode driver.

In other words, 10-bit file with 8-bit rendering was broken, now it's not.

[wanezhiling]

I bought a Surface 3 (not Pro), x7 z8700 is capable of 8-bit hevc, indeed.

http://i3.tietuku.com/c4aaf96029a52ec8.jpg

http://i3.tietuku.com/7d83b4807bd6a620.jpg

http://i3.tietuku.com/6b66755d815712d5.jpg