Evaluation of HEVC decoders (SW, Hybrid and HW)

[P.J]

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

https://www.google.com/search?es_sm=93&q=print+screen+key&spell=1&sa=X&ved=0CBoQvwUoAGoVChMIp8Cnn_SDyAIVKZ3bCh1cOw66

Try to benchmark some of the videos here and post the result.

[wanezhiling]

Try to benchmark some of the videos here and post the result.
Those clips are too slow for me to download... you know there is a 'firewall' in my country..:o

Well I tested one easy 8-bit hevc clip and it was ~6% slower than hd 530, maybe atom is just too weak in some other chains.

[leonccyiu]

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

Oh wow! Even an Atom can decode HEVC 8 bit in 8k, and do 10 bit files work now after being downsampled to 8 bit?

It'd be nice if hdmi 2 was more prevalent, it's been years since it's been available, even displayport would be nice.

I am going a bit off topic here, but since VP9 is google's codec for Youtube which goes up to 4k60.
I've seen on anandtech that the atom has hardware vp9 decoding.
http://www.anandtech.com/show/9219/the-surface-3-review/4
Although I think it's probably just hybrid like on broadwell and skylake, on your dxva checker I don't see vp9.

I don't think a vp9 dxva protocol exists, I am using a pentium g3258 on a z87 mb so broadwell or skylake would be expensive upgrades, whereas an i5-4690k can do 4k60vp9 in software but comes close to maxing the cpu.

Is there any browser such as chrome that can use the hybrid vp9 decode on Intel Graphics?

[NikosD]

Oh wow! Even an Atom can decode HEVC 8 bit in 8k, and do 10 bit files work now after being downsampled to 8 bit?

http://www.anandtech.com/show/9219/the-surface-3-review/4


Back on May of 2015, I had commented that article asking about HEVC.

I copy the comments here:


NikosD - Monday, May 04, 2015 - link
That was a very good chance to test HEVC decoding capability of Cherry Trail.

How fast is it ?
Is it fast enough to decode 4K@60 fps ?

Does it support 10 bit HEVC decoding in hardware ?
REPLY
Brett Howse - Monday, May 04, 2015 - link
This is something I wanted to try out before the review but I ran out of time. I'll get to this sometime this week and give you an update.
REPLY
Ryan Smith - Monday, May 04, 2015 - link
Intel's specs specifically list Cherry Trail at topping out at 4Kp30 (i.e. Level 5).
REPLY
NikosD - Tuesday, May 05, 2015 - link
Thank you both.
REPLY
Brett Howse - Wednesday, May 06, 2015 - link
I tried a 4K60p 8 bit, and it could almost play it back, but 4k30p 10 bit was a no go. But I'm no expert on HEVC that's Ganesh's field.


So it can do (almost) a 4K60p 8 bit.

Atom has 16 EUs and in playback mode that could be an issue for scaling of high resolution clips such as 4K or even worse 8K.

[sneaker_ger]

and do 10 bit files work now after being downsampled to 8 bit?
What is that even supposed to mean?

[NikosD]

Well I tested one easy 8-bit hevc clip and it was ~6% slower than hd 530, maybe atom is just too weak in some other chains.

If you manage to download even one clip, it would be interesting to show your results in both "Decode benchmark" mode and "playback benchmark" mode along with your GPU clock in order to compare with HD 530 using LAV native and DXVA Checker of course.

P.S
Don't forget to disable EVERYTHING in driver's control panel of Windows, because especially for weak GPUs like those iGPUs from Intel, play a significant role in "playback mode" performance.

[P.J]

and do 10 bit files work now after being downsampled to 8 bit?

:confused:

[daWsOn_s]

Do you know if Broadwell-U CPUs can playback 4k hevc with hybrid decoding?

[NikosD]

Yes it does.

Broadwell supports hybrid decoding of 4K HEVC 8bit/10bit and VP9.

[daWsOn_s]

Yes it does.

Broadwell supports hybrid decoding of 4K HEVC 8bit/10bit and VP9.

I meant, is playback smooth? Meaning is there a limit for the bitrate when the hybird decoding won't be enough? I'll be using 8bit 4k 80Mbit.

[NikosD]

I don't have experience on my own and I haven't seen a lot of discussions around this processor.

Broadwell for desktop is almost non existant, but Broadwell-U is an older processor used in laptops and NUC products.

I read that there are several GPU configurations:

28W with GT3, Iris 6100 Graphics (48 Execution Units)
15W with GT3, HD 6000 Graphics (48 Execution Units)
15W with GT2, HD 5500 Graphics (23 Execution Units for low i3, 24 for others)
15W with GT1, HD (Broadwell) Graphics (12 Execution Units)

80 Mbps is a huge bandwidth for hybrid decoding especially for HEVC 4K, but if you aim for 48EUs iGPU model with HD 6000/6100 you could have a chance.

I don't think there is a hardcoded limit for bitrate, but if your GPU can't decode the clip or the CPU can't keep up feeding the GPU by decoding its part, then there is nothing that can help you.

It would be better to find some answers from someone who has actually tried it though.

[daWsOn_s]

I see, I wanted a new laptop with hevc support but skylake machines are much more expensive so I wanted to save going with broadwell...seems like I could get a crippled support...

[leonccyiu]

I meant, is playback smooth? Meaning is there a limit for the bitrate when the hybird decoding won't be enough? I'll be using 8bit 4k 80Mbit.

On Notebookcheck reviews of the Retina 13 and Macbook with Broadwell they briefly report on h.265 decode abilities.

http://www.notebookcheck.net/Apple-MacBook-Pro-Retina-13-Early-2015-Notebook-Review.139621.0.html

For the Retina 13 which uses an i5U with Iris 6100 graphics it plays back smoothly. There is a graph which shows ~37% cpu usage. I think the Iris 6100 is 50% faster than the hd5500.

For the Core M macbook they say a 4k h265 clip at 2.7mbps is a challenge.

[leonccyiu]

https://blogs.gnome.org/rbultje/2015/09/28/vp9-encodingdecoding-performance-vs-hevch-264/

On this page where several decoders are compared for the same quality video, I notice a faster decoder than ffh265 called openhevc. I can't seem to find a player for windows on the net. I believe lav hevc decoder is derived from ffh265? I've probably got this wrong somewhere

[CruNcher]

@NikosD

Can Cyberlinks Decoder still be used i dont get it to connect :( ?

Though PowerDVD 15 itself shows catastrophical Performance results

here is one of my own results currently with VP6 on the GTX 970 though not DXVA and more Realtime Processing (Multithreading Stability Overhead) oriented then a more synthetic benchmark, i try to task every part of the system in a real Workflow scenario (gonna add the Realtime Encoder CUDA, and ASIC next and task them) :)

and slowly highering the decoding complexity on the DSP Engine and adding more Tasks for the OS to handle @ the same time to bring each slowly to it's knees and measure which dies faster ;)


http://i1.sendpic.org/t/zp/zpUqUHt2JSFa8Zmuy6W5HtpFRzU.jpg (http://sendpic.org/view/1/i/hWRpALVaHzBYpeKDjz8pMj5oqtU.png)
http://i2.sendpic.org/t/hb/hbOOATZ7SduB3qmF6324G4Klpto.jpg (http://sendpic.org/view/2/i/pK45vLwHzJL6Qy7PC15hNYN38jZ.png)
http://i1.sendpic.org/t/lR/lR0rzwR2sCVJQAJHhneUnWPDOGF.jpg (http://sendpic.org/view/1/i/uyfN9USX5QYFdmZYVnf72GB6fpQ.png)

It also shows how nicely Firefox Engine has Improved running in the background while testing and doesn't causing noticeable interference's anymore :)




8 Bit results via CUDA (DXVA2 mix)

Lets go insane (it most probably overloads especially the USB 3 Drive, downloading on the same drive @ the same time) ;)

http://i2.sendpic.org/t/vm/vmBxrHHioMGKao6Wdikqxyhi1Uv.jpg (http://sendpic.org/view/2/i/8diVdHsL1rSRGM5wBWddR5qYNDb.png)

Lets go sane :) 2 sane ones even on DX9 renderer are doable

http://i1.sendpic.org/t/3S/3SoCiPOj79WUuMUFLNbnQZliHRU.jpg (http://sendpic.org/view/1/i/goyQPc39YEeSoEEbk5fuEwhf6gO.png)

Streams come from the same USB 3 Hard Drive Realtime Preview test.

http://i2.sendpic.org/t/dC/dC9RpVliGeLEbxaLVuuAE7NlF1T.jpg (http://sendpic.org/view/2/i/65AOb6l1FPkU6zZODWrJYoxJqFD.png)

System = Sandy Bridge I5-2400 on Windows 7 Nvidia WDDM 1.1



Totally unstable playback results PowerDVD 15 the whole thing is still 32 Bit :(

http://i2.sendpic.org/t/yb/ybLD6xcBGS9PhYKagjE0K5dWeC7.jpg (http://sendpic.org/view/2/i/vFNfGe0pYKq0YtYWp3C8YbiDajr.png)

[NikosD]

I see, I wanted a new laptop with hevc support but skylake machines are much more expensive so I wanted to save going with broadwell...seems like I could get a crippled support...

It looks like that the potential buyer of a Core M (Broadwell - Y) should be extra careful of the exact product.

According to anadtech.com, the Intel Core M 5Y10c with iGPU HD 5300 doesn't support hybrid HEVC at all! (not even 8bit)

http://www.anandtech.com/show/8303/zbox-sphere-oi520-plus-zotacs-take-on-the-nuc/5

Of course this kind of CPU, would never find a place in a laptop.

[nussman]

Broadwell and skylake are pretty much useless for HEVC, because no 10bit support.
I would wait or buy something cheap.

1080p 8bit Clips with "normal" bitrates can be decoded by a modern CPU.

[CruNcher]

Seeing Intel lose here i can hardly believe it

It's expected that AMD would take the lead with its more capable graphics core and UVD engine, but by what margin?

Not really

https://youtu.be/lliDV585AT0?t=126

But i guess it's a business decision saving the Space for more important transistors for the current timeframe of content distribution comparable to Nvidias sometimes strange looking decisions on the VPx layer.

They practically even showed that @ IDF they see no need for it in 2015 so why waste space now, it makes perfect sense that AMD though can still compete with UVD and still maintain better 3D Performance says a lot but energy efficiency wise it says not much.

Most Reviewers though never correctly look into these things most of the times ;)

Next year we will see all of them supporting 10 bit like the ARM SOCs do already this year.

Amd is really catching up on Intels Shrinking run if Intel doesn't slowly improves the GPU efficiency i guess it will be a full win for AMD next year maybe even a full Win in 2 areas vs Nvidia and Intel.

https://www.youtube.com/watch?v=9C9x-TB_0Ro

i guess we can believe these PR though @ which TDP ?

https://www.youtube.com/watch?v=Rp1nzQVAZKE

HSA (Fusion) seems to slowly pay off

https://www.youtube.com/watch?v=K6gqBE2_RxM

On the other side see how much load AMD puts on Intels Quicksync 64 Mbps that's far off from a Internet Streaming scenario that would make sense this year even in 8 bit ;)

Pretty well optimized

https://www.youtube.com/watch?v=26kF8vJO3gc

[Yups]

It looks like that the potential buyer of a Core M (Broadwell - Y) should be extra careful of the exact product.

According to anadtech.com, the Intel Core M 5Y10c with iGPU HD 5300 doesn't support hybrid HEVC at all! (not even 8bit)

http://www.anandtech.com/show/8303/zbox-sphere-oi520-plus-zotacs-take-on-the-nuc/5

Of course this kind of CPU, would never find a place in a laptop.


August 1, 2014



Outdated...

[mariush]

...


You can read a lot of technical information about Carrizo and its specs in the Anandtech article: http://www.anandtech.com/show/9319/amd-launches-carrizo-the-laptop-leap-of-efficiency-and-architecture-updates

15w TDP, up to 8.3 hours for h264 playback or 300 minutes for hevc 4k 60fps (with a 50wh battery), lots of power efficiency tweaks...

Page 6 has a lot of details about UVD6 :

http://www.anandtech.com/show/9319/amd-launches-carrizo-the-laptop-leap-of-efficiency-and-architecture-updates/6

It will be up to laptop manufacturers to use these, and that's where amd had problems in the past.. as far as i can see, intel has the monopoly. Maybe since this carrizo has an on-die arm core for security, more business grade laptops will be made with it.

[NikosD]

August 1, 2014



Outdated...

Nice catch, wrong link by me.

The only reason I can think of that, is that I had two tabs open and copied the wrong link.

The right link is here:
http://anandtech.com/show/9689/ecs-liva-core-review-a-fanless-core-m-minipc/5

[Yups]

Nice catch, wrong link by me.

The only reason I can think of that, is that I had two tabs open and copied the wrong link.

The right link is here:
http://anandtech.com/show/9689/ecs-liva-core-review-a-fanless-core-m-minipc/5



I checked the driver thread.....no Hybrid HEVC for Core M is intended: https://communities.intel.com/thread/59216

[CruNcher]

You can read a lot of technical information about Carrizo and its specs in the Anandtech article: http://www.anandtech.com/show/9319/amd-launches-carrizo-the-laptop-leap-of-efficiency-and-architecture-updates

15w TDP, up to 8.3 hours for h264 playback or 300 minutes for hevc 4k 60fps (with a 50wh battery), lots of power efficiency tweaks...

Page 6 has a lot of details about UVD6 :

http://www.anandtech.com/show/9319/amd-launches-carrizo-the-laptop-leap-of-efficiency-and-architecture-updates/6

It will be up to laptop manufacturers to use these, and that's where amd had problems in the past.. as far as i can see, intel has the monopoly. Maybe since this carrizo has an on-die arm core for security, more business grade laptops will be made with it.

Apu Silicon measured a total of 30 Watts for the Laptop he tested.

So they seem to more cram the Performance Gain not into a Power Saving but a Performance Gain especialy on the GPU side

https://www.youtube.com/watch?v=G4YT01OkYOo

Yeah but the Trustzone part seems only to be active in the business laptop part (for now) not the consumer the consumer is the FX-8800p the Business one is another Model.

But the thing is the Trustzone Capability is also in the Consumer version it's just sleeping same as the TPM part.

And you most probably can't change the OS on these Systems you will be locked to Windows 10, Installing Linux for example could become a major problem as Secureboot is Mandatory for Windows 10 upto Windows 8 it was still optional for the Manufactures to get the Windows Logo.

Now you will be forced to enter the new Signed Windows World and give up total control of the System like you do in the Apple world already.

[Xebika]

DXVA Checker v3.9.0 is out

Version 3.9.0 (2015/10/12)

Added the following alternative names of Decoder Device
VP8_VLD
VP9_VLD_Profile0

Version 3.8.0 (09/13/2015)

Now check 8K resolution for Decdoer Device and Processor Device
Changed "4KUHD" to "4K"
Fixed not detect "HEVC_VLD_Main10" when you select "H.265/HEVC" as format even if decoder support it
Minor changes and fixes

[P.J]

DXVA Checker v3.9.0 is out

Version 3.9.0 (2015/10/12)

Added the following alternative names of Decoder Device
VP8_VLD
VP9_VLD_Profile0

Version 3.8.0 (09/13/2015)

Now check 8K resolution for Decdoer Device and Processor Device
Changed "4KUHD" to "4K"
Fixed not detect "HEVC_VLD_Main10" when you select "H.265/HEVC" as format even if decoder support it
Minor changes and fixes

GTX960:
VP9_VLD_Profile0: DXVA2/D3D11, SD / HD / FHD / 4K

[nevcairiel]

Before you get all excited, Microsoft (or NVIDIA, or Intel) still needs to release a DXVA2 specification before anyone can implement a decoder for this.

Edit:
Actually, they did, and I just found it. Guess I have some work to do.

[huhn]

i highly doubt that VP9 gets important.

not part of UBD spec and the google VP9 encoder is nightmare slow.

[baii]

Chrome is pushing VP9 YouTube down everyone's throat. Say if Google buy Netflix or w.e.
i highly doubt that VP9 gets important.

not part of UBD spec and the google VP9 encoder is nightmare slow.


Sent from my 306SH

[huhn]

it's not like you have to play the VP9 version.

[wanezhiling]

Before you get all excited, Microsoft (or NVIDIA, or Intel) still needs to release a DXVA2 specification before anyone can implement a decoder for this.

Edit:
Actually, they did, and I just found it. Guess I have some work to do.
Wow, tha't great, can't wait. :p

ps: currently, only skylake and nvidia gm206 support vp9??

[nevcairiel]

Looks like it doesn't actually work yet on latest NVIDIA drivers. The specification is only 5 days old, so that may not be that big of a surprise. I should boot up my Haswell and see if that exposes it as well.

[NikosD]

I think you need Broadwell or better.

For Haswell is "empty"

[CruNcher]

Guess VP9 support comes with VP7 only like 10 Bit HEVC ?

Does AMD also supports it in the FX-8800p Carrizo UVD, i saw no demonstration of it in their marketing ?

Though neither Nvidia really seemed to advertise anything of it much.

Theoretically Nvidias Android TV Shield Console should also support it 10 bit HEVC + VP9 ?

Hmm those no named only IDs are the PP filter ?

http://i2.sendpic.org/i/e3/e3BCiamvZYeltcax0LMVWzvkMbG.png

[NikosD]

I checked the driver thread...

Using DXVA Checker v3.9.0, does Skylake's drivers provide VP8/VP9 support ?

[pacuro]

This is my first post on Doom9's. Reading here since few years and learning from the best.
Now have something to add and hope it helps.
I have GTX960 so confirm P.J's result regarding VP9 in DXVA Checker.
Additionaly asked my collegues for DXVA Checker 3.9.0 screenshots from i3 Broadwell NUC - HD5500 and i7 Skylake - HD530. Does it look like no VP9 hw support?

http://static.frazpc.pl/board/2015/10/c9cec425f38a5869.jpg

http://static.frazpc.pl/board/2015/10/07c206b1ad06527c.jpg

http://static.frazpc.pl/board/2015/10/b546c9a604df7b53.jpg

[nevcairiel]

The official (as in from Microsoft) VP8 and VP9 DXVA2 support was only officially released last week. Just because its not shown right now doesn't necessarily mean it never will be. Drivers may need to adapt to the final version of the specification first.

[pacuro]

... and another one collegue with Braswell NUC (Pentium N3700 - HD Graphics). Seems to be better, like GTX960.

http://i60.tinypic.com/nej7yq.jpg

[rbej]

No HEVC 10 bit support in Braswell??.

[CruNcher]

Anyone with VP7 (GTX960,GM206) Android Shield TV Console (X1), Maxwell Mobile starting with GTX 940M

can test this clip please on the latest driver

http://demo-uhd3d.com/fiche.php?cat=uhd&id=67



Jesus Samsungs NX1 samples where allready crazy but this is insane

http://demo-uhd3d.com/fiche.php?cat=uhd&id=112

[P.J]

This is my first post on Doom9's. Reading here since few years and learning from the best.
Now have something to add and hope it helps.
I have GTX960 so confirm P.J's result regarding VP9 in DXVA Checker.
Additionaly asked my collegues for DXVA Checker 3.9.0 screenshots from i3 Broadwell NUC - HD5500 and i7 Skylake - HD530. Does it look like no VP9 hw support?

Please test some of the previous clips with them :thanks:


Anyone with VP7 (GTX960,GM206) Android Shield TV Console (X1), Maxwell Mobile starting with GTX 940M

can test this clip please on the latest driver

http://demo-uhd3d.com/fiche.php?cat=uhd&id=67



Jesus Samsungs NX1 samples where allready crazy but this is insane

http://demo-uhd3d.com/fiche.php?cat=uhd&id=112

GTX960 has no issue with any HEVC clip :)

[nevcairiel]

Only 8-bit and only 100mbit? Thats not too bad. =p

If these are clips directly recorded with a camera without re-encoding, then they are usually not very "complex" clips, coding-wise, as cameras just encode with an extremely high bitrate, but not the complex coding tools, because its easier to build them that way.

[CruNcher]

Yeah Pardon it's Sonys XAVC not HEVC so overall the Samsung NX1 samples easily kill it in complexity

[NikosD]

This is my first post on Doom9's. Reading here since few years and learning from the best.
Now have something to add and hope it helps.


Welcome and thank you for your post.

Really helpful.

[NikosD]

I decided to upgrade my ADSL line (14 Mbps/ 1 Mbps) to FTTC+VDSL (50 Mbps / 5 Mbps) and I have to say that it's the first time that I managed to watch in realtime - using Web browsers - 4K youtube clips.

Now, I can upload easily even bigger test files :)

[ashlar42]

My HTPC has a GTX 660, under Windows 8.1 x64. I use Kodi as a media player, hence I believe I'm forced to use x86 versions of LAV Filters. Any point in evaluating this Lentoid sw decoder or is LAV better and capable of partially offloading h.265 decoding to GPU?
Thanks, I searched for Lentoid here but the latest result dated back to April.

Edit: forgot about my CPU, relevant for sw decoding, it's a Q9550S (http://ark.intel.com/products/40815/Intel-Core2-Quad-Processor-Q9550S-12M-Cache-2_83-GHz-1333-MHz-FSB).

[pacuro]

Please test some of the previous clips with them :thanks:

It will be really difficult, as these guys do not care a lot about quality and performace of playback and comfortable watching movies. They have sent me printscreens and it was painfull for them :D
What? MPC-HC? LAV? DXVA Checker? :D

Will try anyway, maybe when visiting them once.

[Yups]

Using DXVA Checker v3.9.0, does Skylake's drivers provide VP8/VP9 support ?


Looks like there is VP9 support on HD 530 (20.19.15.4300).


http://fs5.directupload.net/images/151025/4epk7yr4.jpg

[NikosD]

It's been discovered here in Doom9 by Wanezhiling - not for the first time I think - a serious performance hit for Intel's iGPUs when using NV12 video output (the default option for LAV Video and Microsoft's DXVA) compared to YV12.

You can read more here:
http://forum.doom9.org/showthread.php?p=1744279#post1744279

Now, according to those findings, I opened a thread to Intel's forum hoping for a fix.

The link/thread is here:
https://communities.intel.com/thread/88410

For anyone interested in Intel's iGPUs, you can open an account in Intel forums or use your existing one and add your voice with your experience about the subject, writing at the above Intel thread, in order to find a solution.

Nintendo Maniac 64 has already done that.

Intel has replied nothing till now.

I think we have to "push" them a little more!

Thanks!

[nussman]

Die you try to compare the Videoprocessing (Luma and chroma scaling, Deinterlacing)?
Maybe in YV12 you can not use the adaptiv GPU Deinterlacer (bob/weave Fallback) or the "good" GPU scaler (bilinear fallback).

[clsid]

Also disable all "enhancements". For example Skin Tone Enhancement and Contrast Enhancement.