Intel QuickSync Decoder - HW accelerated FFDShow decoder with video processing [Archive] - Page 50

NikosD

29th April 2014, 17:15

For SandyBridge there was no DXVA HW VC-1/WMV3 decoder, at least before latest driver release.

If it happened to play such file on Sandy, then CPU was used for sure when using DIVA (fallback to CPU decoding)

But QS for HW VC-1/WMV3 on Sandy works fine.

For Sandy and all the other iGPUs the main use of QS is along with a discrete GPU, when dGPU handles 2D, 3D and iGPU handles video decoding.

andyvt

29th April 2014, 19:12

That's impossible!

DXVA always uses as low CPU as possible, in every configuration.

Try it again.

Depending on the CPU it's quite possible for DXVA decoding to take 20% of CPU. The trick however is that CPU throttling could be in effect.

For e.g.
https://lh4.googleusercontent.com/-0ANXVarEDpE/Uzl6i_4fm6I/AAAAAAAAOp8/BfSvuKWxZOM/s640/dxva2.jpg

NikosD

29th April 2014, 19:32

That's different.

The way it was written by iontutm80, it was like DXVAn needs more CPU power/ usage than QS.

Nice catch though, he should check CPU frequency too.

andyvt

29th April 2014, 19:50

...it was like DXVAn needs more CPU power/ usage than QS.

Yes, that's simply not possible because QS = DXVA + copying frames around in memory. On the N2820 this is crippling, probably because it only has one channel of DDR-1066 exposed (Bay Trail supports two, but the board only has one slot). QS and DXVA-CB are both unusable (although QS is worse than DXVA-CB).

My point was that it's quite possible for the facts as they were represented to be correct, it's just the interpretation that could be inaccurate.

ionutm80

30th April 2014, 07:01

That's different.

The way it was written by iontutm80, it was like DXVAn needs more CPU power/ usage than QS.

Nice catch though, he should check CPU frequency too.

Mea culpa. I do not have the SB anymore so unfortunately I cannot check again. But could have been my mistake...

Anyway I have tested the majority of your samples that you use for tests (except those 2160p which none played) with my Celeron NUC + DXVA native in LAV and I can say that everything went very smooth (15% load on CPU in average, peaking to between 23-30% sometimes on some high bitrate movies). Even the "bad ass" Samsung clip went very smooth. I still have some m2ts files taken with a Sony camera (CX-115e), 1080i 50fps that I have not test it yet and which prooved to be very cumbersome in the past. In LAV filters I have noticed that H/W deinterlacing is only possible with QS and not with DXVA native...

Thanks again for your helpfull clarifications and apologize if I said something stupid / not accurate.

NikosD

30th April 2014, 10:27

Anyway I have tested the majority of your samples that you use for tests (except those 2160p which none played) with my Celeron NUC + DXVA native in LAV and I can say that everything went very smooth (15% load on CPU in average, peaking to between 23-30% sometimes on some high bitrate movies).

Your NUC should be play those 4K clips using your iGPU (Ivy)

In order to do that, you should tick the UHD (4K) box in resolutions tab of LAV Video properties.

If you do it, give me a feedback.

In LAV filters I have noticed that H/W deinterlacing is only possible with QS and not with DXVA native...

By using DXVAn and EVR/EVR-CP renderer, HW deinterlacing is automatically used without selecting it anywhere.

ionutm80

30th April 2014, 10:52

Your NUC should be play those 4K clips using your iGPU (Ivy)

In order to do that, you should tick the UHD (4K) box in resolutions tab of LAV Video properties.

If you do it, give me a feedback.

Done that, ticked the UHD box in LAV, tested CrowdRun and OldTownCross, only the OldTownCross played very choppy (however with decent CPU load, ... I don't remember the figure exactly). For CrowdRun I received an error mesage about not possible to render the file. I will download all your 2160p and revert with results.

By using DXVAn and EVR/EVR-CP renderer, HW deinterlacing is automatically used without selecting it anywhere.

Thanks for info, I will test tonight the 1080i 50fps files and revert.

NikosD

30th April 2014, 11:54

Maybe run the DXVAchecker first, in order to be sure that 4K is enabled in drivers for your HW.

andyvt

30th April 2014, 12:11

Your NUC should be play those 4K clips using your iGPU (Ivy)

I wasn't able to get 4K DXVA2 decoding to work properly on the N2820 on a 1080p display (don't have a 4K display to test it). It can decode it, but the results weren't acceptable. My guess is that it's bandwidth limited or that it runs out of go downscaling 4k->1080p.

ionutm80

30th April 2014, 12:33

I wasn't able to get 4K DXVA2 decoding to work properly on the N2820 on a 1080p display.

Did you ran by chance DXVA Checker as Nikos suggested?

NikosD

30th April 2014, 12:35

andyvt

30th April 2014, 12:40

I think it's the second.

The number of EUs is extremely limited (only 4 I think), so the GPU load should be the main restriction factor of 4K decoding and downscaling to 1080p.

Can you check GPU load during 4K decoding on a 1080p display ?

My understanding is that decoding is mostly ASIC, with limited VPP taking place in EU. Has Intel documented where scaling occurs?

IIRC, I was having trouble getting GPU-Z to work on the N2820. Do you know of another tool? That's why I'm hesitant to only blame the GPU (I don't have a way to prove it :)).

NikosD

30th April 2014, 13:15

I use HWiNFO it's very good and detailed and actually I use the HWiNFO monitor, which is a Win 7 gadget in order to see realtime some sensors (including clocks and GPU load of iGPU)

andyvt

30th April 2014, 13:19

I use HWiNFO it's very good and detailed and actually I use the HWiNFO monitor, which is a Win 7 gadget in order to see realtime some sensors (including clocks and GPU load of iGPU)

Thanks. I'll give that a try.

andyvt

30th April 2014, 13:41

can you check gpu load during 4k decoding on a 1080p display ?

93.42%

NikosD

30th April 2014, 14:48

Thanks.

The number of EUs is the limiting factor for proper 4K decoding and downscaling to 1080p, I'm pretty sure after your result.

ionutm80

30th April 2014, 15:42

Thanks.

The number of EUs is the limiting factor for proper 4K decoding and downscaling to 1080p, I'm pretty sure after your result.

Tried CrowdRun on a Intel Core I5 3320M @ 2.60 GHz, HD Graphics 4000, 16 EUs. See attached (UHD checkbox ticked in LAV, QS activated, DXVA similiar at CPU load). Are sure it's related to the number of EUs?

NikosD

30th April 2014, 16:23

I can't see your attachment, but if you have problems decoding 4K with your Core i5 try to put everything in Intel's driver menu from here:
VIDEO - > color enhancement, VIDEO -> Image enhancement, VIDEO -> Image scaling
to "Application settings" and "OFF"

Be sure to do that in your NUC too

theoneofgod

1st May 2014, 16:06

Tried CrowdRun on a Intel Core I5 3220M @ 2.60 GHz. See attached (UHD checkbox ticked in LAV, QS activated, DXVA similiar at CPU load). Are sure it's related to the number of EUs?

QS or DXVA doesn't seem to work for me with MPC and 4k movies. It always uses avcodec and lags like crazy...

ionutm80

5th May 2014, 07:06

Went home and tried the same CrowdRun this time on a desktop Core I3 3220T with HD Graphics 2500. Lags like hell and all 4 virtual cores were at 100%. But the strange thing was that MPC-HC was not displaying "Playing H/W" like in the case of the Core I5 3320M which has HD Graphics 4000!!! Anyway my bet is that downscalling from 2160p to 1080p (I don't have a 4k capable monitor nor a display port available on the MB to test if only video decoding works smooth) is too much for my IVBs not to mention also NUC Atom.

NikosD

5th May 2014, 10:24

The 4 virtual cores running at 100% and MPC-HC not displaying "Playing H/W" as you can easily imagine means bad configuration.

If you use MPC-HC, you must tick 4K box inside MPC-HC and LAV internal filters, which have separate configuration from LAV external filters.

ionutm80

5th May 2014, 10:36

NikosD

5th May 2014, 11:15

Try this http://www.codecguide.com/download_other.htm

It's called Codec Tweak Tool and has an option to fix registry codec problems.

Have you tried any other 4K H. 264 sample ?

ionutm80

5th May 2014, 11:41

Try this http://www.codecguide.com/download_other.htm

It's called Codec Tweak Tool and has an option to fix registry codec problems.

MPC-HC has all the codecs embedded (LAV Filters) so should be no interraction with other codecs packs, no? I have only shark007 installed but the light version with only LAV Filters w/o FFDShow and other.

Have you tried any other 4K H. 264 sample ?

Tried also OldTownCross, same issues. By the way I've left DXVA native and not QS and of-course I have ticked 4k.

P.J

9th May 2014, 10:32

Still no HEVC H.265 support through QuickSync? :(

I had no problem with Intel (HD4400,DxVA/QS), 4K 125Mbps:

http://t.imgbox.com/bKJiuiHr.jpg (http://imgbox.com/bKJiuiHr)

theoneofgod

12th May 2014, 02:46

3770k. Intel 10.18.10.3574. LAV Filters 0.61.2. 4K LG UHDTV Garden Demo.mp4
Distortion at the bottom around 48 seconds. The latest beta drivers helped as it happened at the start.
Not only at 48 seconds by the way.

mp4 can be grabbed here: http://btdigg.org/search?info_hash=afb550e11dd70fab79399f6f430fc8af2c9cfc84&q=4k

NikosD

12th May 2014, 13:14

That 3min clip is huge, 2.69GB, due to 125Mbps rate!

Using latest nightly build of MPC-HC and LAV internal filters, I got a lot more problems using DXVAn than QS.

In both modes though, there is no constant image distortion.

Only in specific points and mainly after seeking.

theoneofgod

12th May 2014, 18:26

Here, setting DXVA reverts to avcodec. Distortion at 3:04 too. The 0:38 distortion happens without seeking, everytime.
Exact same issues with MPC-HC nightly.

nevcairiel

12th May 2014, 19:47

If it also happens with the software decoder (ie. avcodec), then the stream is most likely damaged.

theoneofgod

12th May 2014, 20:53

I can't see any distortion with avcodec.

NikosD

12th May 2014, 20:55

S/W decoding seems fine.

nevcairiel

13th May 2014, 00:25

You are posting confusing things then.

Here, setting DXVA reverts to avcodec. Distortion at 3:04 too. The 0:38 distortion happens without seeking, everytime.

You said DXVA reverts to avcodec, and it shows distortion too? or not? or what now?

theoneofgod

13th May 2014, 10:13

You are posting confusing things then.

You said DXVA reverts to avcodec, and it shows distortion too? or not? or what now?

I was adding to the QS distortion. With QS, distortion also occurs at 3:04, as well as 0:38. I'm still unable to have DXVA function with 4k streams.

M@xim

17th May 2014, 12:19

Where is better picture quality now, DXVA or QS?

kypec

17th May 2014, 12:29

Where is better picture quality now, DXVA or QS?
Picture quality must not be dependent on used decoder. If you experience any visual difference between decoders used then there must be some post-processing going on between decoder output -> renderer input.

Yups

4th June 2014, 13:25

Intel 10.18.10.3621 drivers for Haswell, Ivy Bridge & Bay Trail. Eventhough the page says only for NUC, it does install on all supported hardware.

https://downloadcenter.intel.com/Detail_Desc.aspx?DwnldID=23886

https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProdId=3720&DwnldID=23885&ProductFamily=Grafik&ProductLine=Desktop-Grafikcontroller&ProductProduct=Intel%C2%AE+Core%E2%84%A2+Prozessoren+der+vierten+Generation+mit+Intel%C2%AE+HD-Grafik+4600&lang=eng

Interesting is this change:

HEVC decode now supported by the driver and video players can now take advantage of the GPU accelerated decode support offered by Intel

theoneofgod

4th June 2014, 15:03

https://downloadcenter.intel.com/Detail_Desc.aspx?agr=Y&ProdId=3720&DwnldID=23885&ProductFamily=Grafik&ProductLine=Desktop-Grafikcontroller&ProductProduct=Intel%C2%AE+Core%E2%84%A2+Prozessoren+der+vierten+Generation+mit+Intel%C2%AE+HD-Grafik+4600&lang=eng

Interesting is this change:

If they mean via QS, I see no difference. Same old distortion.

CiNcH

5th June 2014, 11:36

HEVC decode now supported by the driver and video players can now take advantage of the GPU accelerated decode support offered by Intel
Indeed interesting. I wonder how it is done. Since H.265/HEVC QS ASIC support does not even appear on the list of Broadwell, it can only be some acceleration via shaders. It will be interesting to see which parts actually are accelerated and how well it scales with HD2000/HD4000/HD5000 series GPUs.

andyvt

5th June 2014, 11:38

Yups

5th June 2014, 13:34

Indeed interesting. I wonder how it is done. Since H.265/HEVC QS ASIC support does not even appear on the list of Broadwell, it can only be some acceleration via shaders. It will be interesting to see which parts actually are accelerated and how well it scales with HD2000/HD4000/HD5000 series GPUs.

Via Shader is more or less confirmed from Intel.

On that note, there are plans to release a "hybrid" (utilizing EUs of the Core Processor graphics unit) HEVC decoder later this year.
https://software.intel.com/en-us/forums/topic/499189

M@xim

9th June 2014, 02:58

Hi, i used g3420, 10.18.10.3621 video driver, WMP and LAV. When play "Japan 1080p59.94" and use QuickSync in LAV, video play normal, but if i click remote for any position video begins to freeze any ~5 seconds, cpu load in this moments has down to 0. If change QuickSync on DXVA Native video play good.

egur

9th June 2014, 07:11

I'm not familiar with this clip. Can you upload it or part of it?
Which splitter do you use?

M@xim

9th June 2014, 08:13

Splitter is LAV too. Link on download video https://drive.google.com/open?id=0B7zcTvSa9boAUlEyOFluTEdhWGc

egur

9th June 2014, 08:40

Please in a method that doesn't require me to install any SW.

M@xim

9th June 2014, 09:30

Dropbox https://www.dropbox.com/s/aslys4va3ssrsbk/Japan%201080p59.94.m2ts

NikosD

9th June 2014, 09:55

I can confirm the above behavior described by M@xim

Using PotPlayer with QS decoder, seeking becomes almost instant like DXVA native.

theoneofgod

9th June 2014, 11:33

egur

9th June 2014, 12:43

I reproduced also on ffdshow on SandyBridge. I'll look into this.

cybersans

14th June 2014, 12:21

the 15.33.19.64.3540 driver caused out-of-sync while playing mkv/mp4 videos using ffdshow_rev4519_20130622_egur.exe
need to release a new version, maybe since last version is 2013.

fyi, i am using .3412 driver since it is the "latest" driver that did not give any problem with ffdshow. other version later than that until .3540 caused audio and video not sync, where i can see that video comes faster than audio.

thats all for the information.

CharlieCL

19th June 2014, 19:50

Weird true full screen bug. I tested with windows 8.1, LAV 0.61.2 on Dell XPS12 Haswell. The first time installation I can switch to true full screen and run videos fines. But when I switch back to normal windows mode then back to true full screen, the full screen stayed in dead, no any reaction except close it. I am not sure if Quick Sync or driver not release all the resources in full screen switching.