DGAVCDecNV 1.0.13: GPU decoding on Nvidia

[Zwitterion]

Post your script please.
DGAVCDecodeNV_AVCSource("00014.dga")
Does it decode fine for you?

[Sharktooth]

If it's coded in some silly managed language, I'm not going to bother trying to debug it.
i have a gf6800 so i cant help here...

[Guest]

Does it decode fine for you? I can't get file sharing sites behind my corporate firewall, so I can't answer that until this evening.

[Inventive Software]

Yes, it seems to work fine. Why would you be amazed? Allowed by what? I haven't done any comparisons yet.

I'm amazed that it's so simple to implement and NVIDIA would allow something like that. If I'm honest, I thought they'd lock you to their PureVideo de-interlacer.

On a separate note, don't you have an FTP people can post to, or am I thinking of someone else? :)

[Atak_Snajpera]

So a word of warning to Vista users - don't use the two DLLs supplied in the package (nvapi.dll and ncuda.dll) as I think they are from Winxp and must be different to the Vista ones (which are already in system32). Looks like nvcuvid.dll is still needed.
There is simple solution : if OS=6.0 then use system32dlls else use dlls in main direcory

[blubberbirne]

what the hell is going on :devil:

i test the new version with some orf hd test file (1280x720 50P)

First Pass: encoded 3618 frames, 19.61 fps, 4972.53 kb/s
Second Pass: encoded 3618 frames, 87.28 fps, 4994.29 kb/s

i used these settings:


x264.exe --pass 1 --bitrate 5000 --stats ".stats" --level 4.1 --bframes 3 --b-adapt 2 --direct auto --filter -1:-1 --subme 2 --partitions none --vbv-bufsize 50000 --vbv-maxrate 50000 --me dia --threads auto --thread-input --progress --no-psnr --no-ssim --output NUL "dga.avs"
pause
x264.exe --pass 2 --bitrate 5000 --stats ".stats" --level 4.1 --bframes 3 --b-adapt 2 --direct auto --filter -1:-1 --subme 2 --partitions none --vbv-bufsize 50000 --vbv-maxrate 50000 --threads auto --thread-input --progress --no-psnr --no-ssim --output "video.mp4" "dga.avs"
pause

can't belive it, but the encoded video works fine here :thanks:

[TL0]

what the hell is going on :devil:

Your first pass is slower because of frame decision, second pass does not have to do this and is much faster because of your lowered quality settings.

[blubberbirne]

Your first pass is slower because of frame decision, second pass does not have to do this and is much faster because of your lowered quality settings.

normaly i get ~12fps in second pass with these settings

[TL0]

You should maybe ask one of the x264 developers about this then, I looked at your settings and they are almost indentical for both passes, except for --me hex on second pass. :confused:

[DarkZell666]

Simple : you're using --b-adapt 2, which is much slower in the first pass than on the second (since it does all it's magic in the first pass).

[blubberbirne]

hey, the second pass is the problem ;) Its "ultra" fast for HD Encoding ;)

[Guest]

First time I've ever had a complaint that my software is too fast. :)

[AtomX]

I am very happy to report that everything works flawlessly as far as I can find (flawlessly for my workflow anyway). Here's how it goes:

1. Process an M2TS file from a ripped blu-ray (in this case, The Girl Who Leapt Through Time) through DGAVCIndexNV
2. Create an AVS script to process the .dga
3. Open up my Premiere project that uses an MJPEG encode of the movie (made from the blu-ray, originally using the old DGAVCIndex
4. Replace the footage with the new AVS script
5. Everything is in it's right place and works perfectly (granted it's slow, but I mean, it's 1080p. I would never expect it to be)

I can export from Premiere in full 1080p and everything works great. What I had to do before was demux the m2ts file to a .h264 file, process that through DGAVCIndex, and then bring that into VDub to encode it to both MJPEG and to Lagarith (resulting in an enormous file, and the encode time was very significant), and then bait and switch between the two.

So basically, having this new workflow is going to save me a ton of HD space and a lot of time. Thanks a ton! :D

[Ranguvar]

My #1 request: Linux support. There is CUDA for Linux. If any Linux devs are willing to help neuron2, then, since I don't believe he uses Linux at all? :)

[Atak_Snajpera]

Neuron2 - Will you add simple support for .MKV?
It would be great if I could import mkv generated by eac3to directly without demuxing.

[blubberbirne]

oh yes, mkv support would be great. i have many avc records muxed to mkv here.

[Guest]

Where is the spec for the MKV container?

[rack04]

Where is the spec for the MKV container?

Is this (http://www.matroska.org/technical/specs/index.html) what you're looking for?

[nm]

My #1 request: Linux support. There is CUDA for Linux. If any Linux devs are willing to help neuron2, then, since I don't believe he uses Linux at all? :)
Unfortunately Cuda Linux SDK doesn't contain the necessary video decoding functions, probably because the display driver doesn't support VP2 in any way.

[Guest]

Linux talk not welcome here! Please go to the Linux forum. Thank you. :)

[Guest]

The only interlaced material I have is the bonus content in SD from a Blu-ray. I tried to decode it with dgavcdecnv100a3 and got weird artifacts:
http://img225.imageshack.us/img225/8017/artdq7.th.png (http://img225.imageshack.us/my.php?image=artdq7.png)http://img225.imageshack.us/images/thpix.gif (http://g.imageshack.us/thpix.php)
It looks like it decodes the stream, but messed up the lines or something like that.
The indexer shows no such artifacts and I can preview the deinterlacer there.
Here (http://rapidshare.com/files/147441106/00014.m2ts.html) is a sample (first 15 MB of the stream). Can anyone else replicate it? Please re-download alpha 3. I have repaired it. I was not honoring the pitch of the Avisynth YV12PICT. Silly me.

Thank you for pointing this out!

http://neuron2.net/dgavcdec/dgavcdecnv100a3.zip

[Snowknight26]

You might like to update your first post/title too.

[Guest]

You might like to update your first post/title too. Great idea, Snowknight26, thank you! I've done it.

[squid_80]

My #1 request: Linux support. There is CUDA for Linux. If any Linux devs are willing to help neuron2, then, since I don't believe he uses Linux at all? :)The cuda video decoder only works with direct3d at the moment. OpenGL support is promised for the future.

[hklck516]

this version can support MeGUI ?

using Vista x64 SP1

[ReferenceDivx]

Great work neuron2! Its people like you who motivate me to do great work.

My setup consists of a q6600 cpu, 8500gt, and 2gb of memory. Tests where run on a raid array.
I'm using Vista 32bits and such got rid of nvapi.dll and nvcuda.dll

compared vs ffdshow filter (my usual method).

Video is 1080p from a bluray with no post processing .

C:\Program Files\megui\tools\x264>x264.exe --pass 1 --bitrate 10345 --stats "tes
t.stats" --level 4.1 --keyint 14 --min-keyint 2 --bframes 2 --direct auto --filt
er -1:-1 --subme 2 --partitions none --ipratio 1.1 --pbratio 1.1 --vbv-bufsize 1
4475 --vbv-maxrate 24000 --qcomp 0.5 --me dia --merange 12 --threads auto --thre
ad-input --progress --no-psnr --no-ssim --output NUL "E:\output\test\test4.avs"
--mvrange 5
avis [info]: 1920x1080 @ 23.98 fps (3431 frames)
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 Cache64
x264 [info]: slice I:328 Avg QP:12.01 size: 86392
x264 [info]: slice P:1900 Avg QP:12.93 size: 67376
x264 [info]: slice B:1203 Avg QP:14.08 size: 33267
x264 [info]: consecutive B-frames: 35.1% 27.0% 37.9%
x264 [info]: mb I I16..4: 70.0% 0.0% 30.0%
x264 [info]: mb P I16..4: 38.0% 0.0% 0.0% P16..4: 19.2% 0.0% 0.0% 0.0% 0
.0% skip:42.9%
x264 [info]: mb B I16..4: 6.2% 0.0% 0.0% B16..8: 18.9% 0.0% 0.0% direct:
21.7% skip:53.2% L0:32.7% L1:38.5% BI:28.8%
x264 [info]: final ratefactor: 17.07
x264 [info]: direct mvs spatial:93.6% temporal:6.4%
x264 [info]: kb/s:10978.1
encoded 3431 frames, 17.10 fps, 10980.39 kb/s

first pass went from 14.04 to 17.10 a 22% increase


C:\Program Files\megui\tools\x264>x264.exe --pass 2 --bitrate 10345 --stats "tes
t.stats" --level 4.1 --keyint 14 --min-keyint 2 --bframes 2 --direct auto --filt
er -1:-1 --subme 2 --partitions none --ipratio 1.1 --pbratio 1.1 --vbv-bufsize 1
4475 --vbv-maxrate 24000 --qcomp 0.5 --merange 12 --threads auto --thread-input
--progress --no-psnr --no-ssim --output "E:\output\test\test.264" "E:\output\tes
t\test4.avs" --mvrange 511 --aud --nal-hrd
avis [info]: 1920x1080 @ 23.98 fps (3431 frames)
x264 [info]: using cpu capabilities: MMX2 SSE2Fast SSSE3 Cache64
x264 [info]: slice I:328 Avg QP:12.43 size: 78128
x264 [info]: slice P:1900 Avg QP:13.51 size: 61252
x264 [info]: slice B:1203 Avg QP:13.97 size: 35959
x264 [info]: consecutive B-frames: 35.1% 27.0% 37.9%
x264 [info]: mb I I16..4: 70.1% 0.0% 29.9%
x264 [info]: mb P I16..4: 35.1% 0.0% 0.0% P16..4: 21.9% 0.0% 0.0% 0.0% 0
.0% skip:43.0%
x264 [info]: mb B I16..4: 6.3% 0.0% 0.0% B16..8: 20.3% 0.0% 0.0% direct:
22.5% skip:51.0% L0:28.8% L1:34.0% BI:37.2%
x264 [info]: direct mvs spatial:86.1% temporal:13.9%
x264 [info]: kb/s:10357.0
encoded 3431 frames, 22.07 fps, 10363.72 kb/s

second pass went from 11.77 to 22.07 a 88% increase

[Inventive Software]

this version can support MeGUI ?

using Vista x64 SP1

MeGUI is NOT supported. Read the whole thread to find out why.:rolleyes:

EDIT: Donald! Matroska specifications: http://www.matroska.org/technical/index.html. Also check out http://haali.net/mkv.

[Guest]

this version can support MeGUI ?
Your error appears to indicate that you aren't finding the DLLs. Make sure that nvcuvid.dll is in the right directory for Vista 64. It's been mentioned several times in the thread how to get things working in Vista 64.

But anyway, once you find the DLLs, MeGui will crash. I need to post in the MeGui development thread about that.

[Sagekilla]

@Inventive Software, the latest version does in fact support MeGUI I believe. neuron2 finished the multi instancing so you can use it properly with MeGUI.

[Inventive Software]

I was under the impression that MeGUI still had quirks that prevented it from working properly. :confused:

[Sagekilla]

The quirks were the fact that MeGUI opens a file several times to check for various parameters. In the case of AVCSource(), it didn't like being called multiple times before.

I believe this was changed, but I could be wrong.

[Guest]

@Inventive Software, the latest version does in fact support MeGUI I believe. neuron2 finished the multi instancing so you can use it properly with MeGUI. My post right above yours says it doesn't work!

I'll repeat it all again:

Yes, AVCSource() now supports multi-instance. But MeGUI does not play nice and crashes. All other apps I have tried work fine. I don't know why MeGUI crashes and I don't plan to try to debug it. Someone needs to report it to the MeGUI developers.

[Sagekilla]

Apologies, I missed your post when I posted.

[Inventive Software]

My post right above yours says it doesn't work!

So I'm not going nutso then! Good to know. ;)

[Sharktooth]

im the only "active" dev... but i already said i cant debug coz i havent a CUDA enabled videocard.

[Guest]

What's MEGUI written in and how hard would it be for me to get a build environment up and running?

[Sharktooth]

c#... managed code... exactly what you hate...

[Turtleggjp]

+1 vote for MKV support, since I'll be using that for all my TV shows this season. I hope you will bring it to the non-NVIDIA version as well, so my laptop can use it too.

Also, you said that you would be able to support VC-1 as well. Will this be fairly simple to add, or will you need to go through a lot of the trial and error stuff you went through getting DGAVCIndex as far along as it is today? I know the decoding part will be easily handled by the CUDA drivers, but I know there's a lot that goes into the indexing as well. No big rush as far as I'm concerned, since the only place I get VC-1 files is from Blu Rays, and I usually don't need to do frame accurate editing on those. Thanks again for all your hard work!

Matt

[Atak_Snajpera]

Also, you said that you would be able to support VC-1 as well.
Do nvidia's cards support VC-1 decoding???

[Guest]

Do nvidia's cards support VC-1 decoding??? Yes, of course.

The VC-1 is defined as an available codec in the cuda header files, too. I haven't tried it yet but will soon.

Theoretically, the CUDA interface is the same for all video types, so if it works it should not be hard to integrate.

[turbojet]

nice work, I have just one little issue that may or may not be resolvable. I often use radmin 3.2 between the 2 computers I have at home. It seems like when I disconnect from radmin it freezes x264 encoding if decoded with dgavcdecnv. Radmin (http://www.radmin.com) has a 30 day trial period I'd really appreciate it if it could be looked at.

[LaTo]

nice work, I have just one little issue that may or may not be resolvable. I often use radmin 3.2 between the 2 computers I have at home. It seems like when I disconnect from radmin it freezes x264 encoding if decoded with dgavcdecnv. Radmin (http://www.radmin.com) has a 30 day trial period I'd really appreciate it if it could be looked at.

why using radmin? tightVNC is free and really good...

[turbojet]

Well I use it at work regularly, and I was able to obtain some licenses from work. They bought a bigger package then they needed, but it didnt cost them any extra, volume pricing. So I didn't actually have to pay for it. I've never had much luck with vnc on windows, slower then im used to and random disconnects, especially when more than one is connected. Even on linux I choose nx over vnc if its for long amounts of time.

[Comatose]

Probably because it uses a virtual video driver or something and it screws with NVIDIA's driver. Remote Desktop does this too.

[lucassp]

Do nvidia's cards support VC-1 decoding???

it seems like only a low end chip has full VC-1 Decoding Support over DXVA:

http://forums.nvidia.com/lofiversion/index.php?t74108.html

[ndkamal]

Hello Neuron2, you make a FANTASTIC JOB, with your software, a graphic card like the GeForce 9500GT (70 €) is more powerful than two C2Q Q9300 (400 €) in transcoding. Thanks a lot !!!!

[Guest]

Don't thank me too hard yet. I have to tell you that for the NV versions I am going to a model like jdobbs has for DVD Rebuilder Pro. That's the reason that the current binaries have an expiration. I plan to roll it out before they expire.

[ndkamal]

A question for blubberbirne what is your processor and what is the fps in second pass with this ?

[rack04]

Don't thank me too hard yet. I have to tell you that for the NV versions I am going to a model like jdobbs has for DVD Rebuilder Pro. That's the reason that the current binaries have an expiration. I plan to roll it out before they expire.

Looking forward to it.

[DarkZell666]

hey, the second pass is the problem ;) Its "ultra" fast for HD Encoding ;)

You used --subme 2 and no partitions, so 80~fps on 720p looks quite normal if you have a quad-core machine :)

Seriously, do compare with --b-adapt 1, I believe it's not the second pass that's too fast but the first pass that's slown down ^^' (though you only used 3 b-frames, hmm wierd).

Please post some test results if you want this settled :)