Evaluation of HEVC decoders (SW, Hybrid and HW)

[NikosD]

I have started uploading there too.

[nussman]

Not so bad at all. Thanks for testing.

[Yups]

Good speed.

4K60 FPS HEVC 10bit at 50 Mbps but with the minimum values below 50fps.

Better than I thought.

What about this one ?

http://demo-uhd3d.com/fiche.php?cat=uhd&id=46

It's from Satellite.


Hybrid

Decode= 64-77-117 [16-25]
Playback= 49-60-93 [13-21%]

Decoding is slightly faster in this lower bitrate video, playback roughly the same. Does a GTX 970 support Hybrid 10 bit decoding?

[NikosD]

Does a GTX 970 support Hybrid 10 bit decoding?

I am not sure.

You have to run DXVA Checker and look for HEVC_VLD_Main10 in the first page.

[nevcairiel]

Does a GTX 970 support Hybrid 10 bit decoding?

No, it does not.

[nussman]

GTX 970 is 8bit hybrid only imho.

[NikosD]

970 would definitely be faster than HD 530 in hybrid 10bit HEVC

[Yups]

It isn't support unfortunately. I will try it next time to make sure though.

[NikosD]

Mega allows even big files like 10gb. Got filedropper to work.

After fully uploading the file (it took me 1 hour) I got a message that in order to create a link to share my upload I had to create an account.

It's a free account, but I will not register to a site just to share an uploaded a file.

[P.J]

After fully uploading the file (it took me 1 hour) I got a message that in order to create a link to share my upload I had to create an account.

It's a free account, but I will not register to a site just to share an uploaded a file.

I think the 300Mbps video I have downloaded from filedropper is corrupted, 360MB.

Ok, I'll wait for new link. Mega has ability to resume upload too.

[NikosD]

No, you misunderstood.

My upload finished OK, but I have no option to share the link.

Where is the link ?

You have to make an account on that site, in order to allow you to share the upload.

You can check the integrity of the file you downloaded from filedropper by running 7 - zip or any other program capable of calculating the checksum of the file.

Run 7 - zip with CRC-32 or SHA-1 and post the number here in order to compare it with mine.

[NikosD]

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.

Your CPU results are very fast.

Using LAV x64 with the default settings of Thread to Multi-Threading = Auto I got these results.

Win 10 x64 - Core i7 4790@3.8GHz - DDR3 1600MHz - LAV x64 0.65.47

http://demo-uhd3d.com/fiche.php?cat=uhd&id=97

Decode 52/62/86 CPU avg 91% ~56W CPU package (HWiNFO)
Playback 40/49/58 CPU avg 83% ~56W CPU package (HWiNFO)

Even if I could run Haswell 4790 at 4.0GHz, our difference would be 52% faster for Skylake !

Your fast RAM could be a reason, but I think it's too much even with that.

Which is your setting for threads of LAV x64 and could you benchmark the ASTRA clip too, posting the average of CPU utilization and the setting of threads for LAV ?

Also, how do you measure the Power consumption and for what components exactly ?

It's two and a half times more than mine !!

My CPU voltage is 1.03 V.

[P.J]

No, you misunderstood.

My upload finished OK, but I have no option to share the link.

Where is the link ?

You have to make an account on that site, in order to allow you to share the upload.

You can check the integrity of the file you downloaded from filedropper by running 7 - zip or any other program capable of calculating the checksum of the file.

Run 7 - zip with CRC-32 or SHA-1 and post the number here in order to compare it with mine.

http://3.t.imgbox.com/JGn5mTT7.jpg (http://imgbox.com/JGn5mTT7)

You can create a fake email. Sendspace is good too.

[NikosD]

You are right that your download is wrong.

The CRC32 is 81DA8738

I tried first sendspace but doesn't allow 360 MB upload.

Could you try once more from filedropper ?

How did you manage to download the other two files ?

I'm open to other servers too but without registration and with easy upload method.

@Yups

Did you have problems too downloading the files ?

Is the 300 Mbps file's CRC32 the right one ?

[P.J]

Someone has uploaded it on mega for me, downloading right now:
https://mega.nz/#!RVNy0BhB!DpsJlf5n_TJV1lptFGH4mAZblTA9-vT0P12VikS3xko

Those two other videos were on sendspace not filedropper.

[NikosD]

So ?

Did you do the tests ?

The results of 300 Mbps file will be interesting, I think.

[Yups]

Your CPU results are very fast.

Using LAV x64 with the default settings of Thread to Multi-Threading = Auto I got these results.

Win 10 x64 - Core i7 4790@3.8GHz - DDR3 1600MHz - LAV x64 0.65.47

http://demo-uhd3d.com/fiche.php?cat=uhd&id=97

Decode 52/62/86 CPU avg 91% ~56W CPU package (HWiNFO)
Playback 40/49/58 CPU avg 83% ~56W CPU package (HWiNFO)

Even if I could run Haswell 4790 at 4.0GHz, our difference would be 52% faster for Skylake !

Your fast RAM could be a reason, but I think it's too much even with that.


This is a big difference indeed. Impossible to say how much the memory speed affects it, maybe I will re-test with DDR4-2133 later this week.



Which is your setting for threads of LAV x64 and could you benchmark the ASTRA clip too, posting the average of CPU utilization and the setting of threads for LAV ?

http://abload.de/img/8kxrav.png

Astra Clip:

Decode= 96-114-171 avg 72%
Playback= 66-84-87 avg 63%


Looks like CPU only handles a lower bitrate video much better. In particular CPU load is lower.


Also, how do you measure the Power consumption and for what components exactly ?


With Energy Check 3000 from the entire system.

[P.J]

So ?

Did you do the tests ?

The results of 300 Mbps file will be interesting, I think.

http://forum.doom9.org/showpost.php?p=1735592&postcount=710

[NikosD]

OK.

Here we are with all clips tested between Skylake HD 530 vs Nvidia GTX 960 using LAV x64 - DXVA native - Win 10 x64 in Decode and Playback (1280 x 720) Benchmark mode using DXVA Checker v3.60 of course (as always).


I have to thank a lot, two members of the forum, Yups and P.J for testing their hardware and giving an amount of their time for the benchmarks.

I just copied their results !

Clips:

1) LG_4K_View-the-Feeling.mp4 - 4K@30fps - 25Mbps
http://demo-uhd3d.com/fiche.php?cat=uhd&id=101

Decode:
Skylake HD 530 154/178/204
Nvidia GTX 960 145/149 /161

Playback:
Skylake HD 530 146/168/202
Nvidia GTX 960 143/148/158


2) Samsung 4K@30fps-80Mbps file -0230
http://www.imaging-resource.com/PRODS/samsung-nx1/FULLRES/yvid_SAM_0230-ProQuality.MP4


Decode:
Skylake HD 530 160/164/168
Nvidia GTX 960 117/118/120


Playback:
Skylake HD 530 147/155/160
Nvidia GTX 960 114/115/118



3) Samsung 4K@30fps-80Mbps file -0235
http://imaging-resource.com/PRODS/samsung-nx1/FULLRES/yvid_SAM_0235-UHD_Sample1.MP4


Decode:
Skylake HD 530 194/202/204
Nvidia GTX 960 119/122/126

Playback:
Skylake HD 530 174/190/201
Nvidia GTX 960 117/118/121



4) Crowd_run 4K@50fps -100Mbps file
https://www.sendspace.com/file/fvgk57


Decode:
Skylake HD 530 160/162/164
Nvidia GTX 960 129/132/134


Playback:
Skylake HD 530 143/151/164
Nvidia GTX 960 128/131/133


5) Crowd_run 4K@50fps -200Mbps file
https://www.sendspace.com/file/usyase


Decode:
Skylake HD 530 133/137/138
Nvidia GTX 960 104/107/109


Playback:
Skylake HD 530 120/129/132
Nvidia GTX 960 104/107/109


6) Crowd_run 4K@50fps -300Mbps file
http://www.filedropper.com/crowdrun2160p50300mbps
or
https://mega.nz/#!RVNy0BhB!DpsJlf5n_TJV1lptFGH4mAZblTA9-vT0P12VikS3xko


Decode:
Skylake HD 530 112/120/123
Nvidia GTX 960 83/93/97


Playback:
Skylake HD 530 102/114/119
Nvidia GTX 960 81/92/97


Comments:

1) There is a clear winner, which is Skylake in both pure Decode mode and Playback too, but with less distance for the latter.

2) Both VPUs inside GPUs, are more than capable of 4K@120fps using 100Mbps HEVC 8bit clips.

3) Samsung clips, especially the second one, favors a lot the fixed - function unit of Intel HD 530 (70% faster than Nvidia).
I don't know how common that could be, with other HEVC 8bit encodings.

4) Skylake HD 530 is close to 4K@120fps even for 300 Mbps clips. It is definitely designed for high bandwidth clips.

[NikosD]

@Yups (more) and P.J (a little)

Did you disabled ALL driver enhancements regarding video in driver's control panel ?

I always do that because especially for Intel iGPUs those enhancements add significant GPU load producing worse image most of the times and drop performance.

It's the first thing I do after installing the driver.

[P.J]

I haven't noticed any settings, which ones exactly?

Only manual settings will reduce the performance like Noise reduction/Edge enhancement...

[NikosD]

Yes, all those exactly.

They are enabled by default in Intel's drivers, I don't remember about Nvidia.

[Yups]

I didn't enable/disable anything in the driver. This is tested out of the box with Intels default settings. I will take a look later for Control Panel settings.

edit: A couple of video settings like noise reduction were enabled. Playback runs indeed faster with all settings disabled/application controlled.

[NikosD]

edit: A couple of video settings like noise reduction were enabled. Playback runs indeed faster with all settings disabled/application controlled.

I had reported that situation a few times in Intel forums and the answer was that the devs know what they are doing and they will keep enabling those features by default in their drivers.

Intel has the least powerful GPUs from all OEMs, but they keep on enabling whatever they want.

Are the results significantly faster with everything disabled or just a couple of frames ?

[NikosD]

Astonishing performance on the first DX12 game benchmark from AMD cards.

They were designed since CGN 1.0 with asynchronous compute engines (ACE), which truly shine using DX12.
On the other hand, all architecture designs of Nvidia cards before Maxwell were based on DX11.
But even the fastest Maxwell card - 980 Ti - is matched in performance by a previous generation AMD card, R9 290X!

While we are still waiting for the DX12 Win10 driver for Fermi, take a look at this:
http://arstechnica.com/gaming/2015/08/directx-12-tested-an-early-win-for-amd-and-disappointment-for-nvidia/

[P.J]

Wish Skylake had 10bit support. What exactly is Intel ClearVideo?

It's soon to comment about DX12, need more new games.

[huhn]

mantle and DX12 have a lot in common.

the main feature is the low level API.

and don't take one test to serious.

[NikosD]

What exactly is Intel ClearVideo?



Intel's ClearVideo is mainly a general term Intel used for all its video processing technologies (SW and HW).

Like PureVideo of Nvidia.

But it depends on the content of the whole concept that the term was mentioned.

Where did you see that ?



It's soon to comment about DX12, need more new games.

mantle and DX12 have a lot in common.

the main feature is the low level API.

and don't take one test to serious.

Of course is early to think about final conclusions regarding DX12 and graphics card's performance, but until we see some major redesign of Nvidia architecture - that was optimized for DX10 and DX11 - for DX12, the numerous SW teams of Nvidia can't help that much with driver's optimizations especially for pre-Maxwell cards.

You know, there is a rumour/ fact floating around that the SW developers inside Nvidia are more than HW engineers, but still they can't do much to transform the HW via SW optimizations.

[huhn]

and it is no secret that an AMD card for the same money has more raw processing power than an nvidia card.

a r9 270 and a gtx 960 have about the same gflops as raw processing power a r9 270 is way cheaper and in games the gtx 960 will clearly win.

there are other test that show that AMD gains more but not as ridiculous as this and more important they show that nvidia is gaining a lot too.

http://www.golem.de/news/vorabtest-directx-12-macht-was-es-soll-1502-112228.html

i would love to see AMD gain some market share but no they didn't add 10 bit HEVC decoding and they didn't add HDMI 2.0 in there new GPUs...

[NikosD]

i would love to see AMD gain some market share but no they didn't add 10 bit HEVC decoding and they didn't add HDMI 2.0 in there new GPUs...

And Intel did the exact same thing like AMD and the new Skylakes.

So, they probably both thinking that 4K60 and BluRay UHD are not very popular yet or even existent.

[huhn]

but people need HDMI 2.0 right now...

[P.J]

Intel's ClearVideo is mainly a general term Intel used for all its video processing technologies (SW and HW).

Like PureVideo of Nvidia.

But it depends on the content of the whole concept that the term was mentioned.

Where did you see that ?

Does ClearVideo support pixel adaptive deinterlacing?
What about Celeron/Pentium series that don't have ClearVideo?

[huhn]

there are some celeron and pentium without clearvideo that didn't have hardware decoder.

deinterlacing should be supported by all but not 100 % sure.

[NikosD]

ALL Celeron/Pentium processors have fixed-function HW decoders and Haswell's Pentium/ Celeron have full QuickSync support, meaning the encode path too.

And they all have adaptive deinterlacing of course.

[huhn]

sandy had the problem but haswell and newer shouldn't be affected anymore.

http://semiaccurate.com/2011/03/21/intels-sandy-bridge-celerons-video-features-fused-off/

Some of the missing features such as InTRU 3D and support for WiDi is sort of understandable on a budget CPU like this, but not only has Intel dropped support for Clear Video HD, but even standard Clear Video isn’t supported which means that the graphics core can’t offload the CPU when it comes to video playback, features you’d expect to be part of an entry level CPU as this is where processor offloading makes the most sense.

[NikosD]

It's dead wrong.

ALL Pentium/Celeron including Sandy have the fixed-function decoder enabled.

[nevcairiel]

sandy had the problem but haswell and newer shouldn't be affected anymore.

http://semiaccurate.com/2011/03/21/intels-sandy-bridge-celerons-video-features-fused-off/

You should read the update to the story, Intel only removed QuickSync on those, decoding is supported. Some of the fixed-function video enhancements have also been disabled, but noone wants those anyway!

[huhn]

good to know.

this was actually an non existing issue people talked about in the past.

so that leaves only the first ATOM with no hardware decoder.

[Yups]

Retested with all image enhancements disabled in the driver.



1) Samsung 4K@30fps-80Mbps file -0230
http://www.imaging-resource.com/PRODS/samsung-nx1/FULLRES/yvid_SAM_0230-ProQuality.MP4

Decode= 160-164-168 [2-3%]
Playback= 147-155-160 [3%]



2) Samsung 4K@30fps-80Mbps file -0235
http://imaging-resource.com/PRODS/samsung-nx1/FULLRES/yvid_SAM_0235-UHD_Sample1.MP4

Decode= 194-202-204 [2%]
Playback= 174-190-201 [2%]


3) Crowd_run 4K@50fps -100Mbps file
https://www.sendspace.com/file/fvgk57

Decode= 160-162-164 [2%]
Playback= 143-151-164 [3%]


4) Crowd_run 4K@50fps -200Mbps file
https://www.sendspace.com/file/usyase

Decode= 133-137-138 [2%]
Playback= 120-129-132 [4%]



5) Crowd_run 4K@50fps -300Mbps file
http://www.filedropper.com/crowdrun2160p50300mbps


Decode= 112-120-123 [3%]
Playback= 102-114-119 [4%]



Biggest difference for video 2. Playback closed in. Decode is slightly slower than my previous test. I did install Bitdefender, maybe it slows down the test a bit.

[NikosD]

Thanks!

I'll update the overall post above.

I'm pretty sure that 10bit hybrid results will be better too.

I'm very interested in your RAM downgrade results to 2133 MHz (default for Skylake) to see how lower could be in SW decoding and maybe HW too compared to 3000 MHz.

[P.J]

Thanks!

I'll update the overall post above.

I'm pretty sure that 10bit hybrid results will be better too.

I'm very interested in your RAM downgrade results to 2133 MHz (default for Skylake) to see how lower could be in SW decoding and maybe HW too compared to 3000 MHz.

Yes, those would be great to test + No Hyperthreading :)
I wonder if i5 6600K can handle 10bit well too.

[NikosD]

Retested with all image enhancements disabled in the driver.


Can you retest this too, in order to complete the picture:

1) LG_4K_View-the-Feeling.mp4 - 4K@30fps - 25Mbps
http://demo-uhd3d.com/fiche.php?cat=uhd&id=101

[Yups]

LG_4K_View-the-Feeling.mp4

decode= 154-178-204 3%
playback= 146-168-202 4%

[Yups]

LG_4K_View-the-Feeling.mp4

decode= 154-178-204 3%
playback= 146-168-202 4%


After a bios update:


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

[NikosD]

The platform is very young and those fluctuations are expected.

I'll keep the previous results of clip 1 in order to be inline with the others.

Unless you want to rerun the tests for the rest too or you tell me that the others didn't change.

[NikosD]

i7-6700k HD 530 1.15 Ghz
10.18.15.4274
Windows 10 x64


@Yups

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'm interested in Intel MFT VP8/VP9 decoders that using previous DXVA Checker v3.60 were enumerated by choosing VP8 and VP9 formats.

Also, if you have a VP8 or even VP9 sample (any sample) can you check with your system and latest DXVA Checker v3.70, if any Intel MFT decoders appear when you choose "Select Media File..." and test a VP8/VP9 file ?

Thanks!

[P.J]

Are upcoming 4K Bluray discs 10bit or 8bit?

[sneaker_ger]

10 bit (UltraHD TV and UltraHD streaming are also often 10 bit already)

[Yups]

@Yups

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'm interested in Intel MFT VP8/VP9 decoders that using previous DXVA Checker v3.60 were enumerated by choosing VP8 and VP9 formats.

Also, if you have a VP8 or even VP9 sample (any sample) can you check with your system and latest DXVA Checker v3.70, if any Intel MFT decoders appear when you choose "Select Media File..." and test a VP8/VP9 file ?

Thanks!


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


http://abload.de/img/unbenanntjbj9p.png

[NikosD]

But the image shows clearly Intel's VP9 decoder :)

If you select VP8 format, does Intel's VP8 MFT decoder appears ?

I'll upload two samples later, one VP8 and one VP9 to test the "Select Media File"