DG NV tools

[rack04]

Yes, it was updated to remove the dependency on D3D.

Please report your OS.

You can continue to use the old one until we resolve this.

Windows XP SP3. I will continue to use CUVIDServer 1.04.

[Guest]

I have XP SP3 too and have no problem.

Let's see what others report.

What is your graphics adapter?

[wangyi6854]

Hi, neuron2.

I also got the error with beta5.

My system is windows server 2008 64bit. Graphics card is g8400gs

[Guest]

Darn. It must be because I applied the Visual Studio service pack.

Investigating...

[wangyi6854]

Another question:

Why force film option is disabled in beta4? I need it to process vob files.

[Guest]

Try installing this:

http://www.microsoft.com/DOWNLOADS/details.aspx?familyid=32BC1BEE-A3F9-4C13-9C99-220B62A191EE&displaylang=en

Force Film is not yet implemented.

[wangyi6854]

It doesn't solve the problem.

BTW: I've installed vs2008 and windows sdk.

[kemuri-_9]

Try installing this:

http://www.microsoft.com/DOWNLOADS/details.aspx?familyid=32BC1BEE-A3F9-4C13-9C99-220B62A191EE&displaylang=en

I'm on XP x64 SP2, and also getting errors running the new version,
i already have the msvc++ redistributables/runtimes for 2005, 2008, and 2010 all installed (both x86 and x64 for each release)...

[Revgen]

Good News: deinterlace=2 works now.

Bad News: I get the same error with CUVIDServer using Winxp SP3, 190.38 drivers, and 8400GS card.

I'll try the VC Redrist package later and see if it works.

[Revgen]

Just tried the package. CUVIDServer still doesn't work. Using the old version instead.

[SomeJoe]

http://forum.doom9.org/showthread.php?p=1310386#post1310386

Sorry, missed that one ...

@neuron2:

Thanks for the new version. Downloading and testing now ...

[SomeJoe]

Finished my tests. Deinterlace=2 now working perfectly, thank you.

I'm in an interesting situation -- I have two identical machines, both are quad core Q6600's running Windows XP SP3 with NVidia 9500GT cards. The new CUVID server 1.0.5 runs perfectly on one machine, but issues the same error that others are getting on the other machine.

The difference is that Visual Studio 2005 with all service packs and patches is installed on one machine, the other does not have it.

CUVID Server 1.0.5 runs fine on the machine with Visual Studio 2005 installed, the error message appears on the other machine.

Currently using CUVID Server 1.0.4 on both machines as a work-around.

[Guest]

Thanks. With all those clues I should be able to find out what is wrong.

[Guest]

Please re-download NV beta 5 and test the server again. I think it should be fixed.

Coming next: resizing.

[kemuri-_9]

Please re-download NV beta 5 and test the server again. I think it should be fixed.

it now works here, what was the problem exactly?

[Guest]

Somehow, the project settings got changed to "multithreaded DLL" instead of "multithreaded". I have no idea how that happened. :confused:

[Guest]

That's old-speak for "video card".

Anyway, never mind that. It apparently was a problem with my project settings. Is the fixed version working for you now?

[rack04]

That's old-speak for "video card".

Anyway, never mind that. It apparently was a problem with my project settings. Is the fixed version working for you now?

I deleted my post since I didn't read all the way through the tread to see that you already fix it.

Yes it works great now. Thanks!

[GZZ]

Coming next: resizing.

The best money spent on a tool in a LONG time. :)

Hardware resizing will properly help alot on the speed (when going 1920x1080 to 1280x720) ??

[Revgen]

CUVIDServer works now. Thanks.

[SomeJoe]

Please re-download NV beta 5 and test the server again. I think it should be fixed.

Tested new Beta 5, now working on both machines no problems.

Coming next: resizing.

Hardware resize would be awesome.

Does NVidia expose a hardware IVTC in their API? Decomb package Telecide/Decimate works very well but is slow with HD material because it runs single-threaded within AviSynth. (I have not messed with MT AviSynth, so perhaps it could be sped up using that, but I haven't tried).

It's no big deal if they don't - I have only needed to IVTC HD material a very few times.

[Guest]

IVTC for hard telecine is not available.

[Guest]

Here is integrated GPU resizing:

http://neuron2.net/dgavcdecnv/NVbeta6.zip

Example for conversion of 1080i29.97 to 720p59.94:

DGSource("file.dgm",deinterlace=2,resize_w=1280,resize_h=720)

[Varies]

resizing without cropping? not sure o_O

what algorithm is used to resize? hardware? its better than spline\lanczos\blackman?

[Guest]

resizing without cropping? not sure o_O I'll add a source cropping rectangle.

I just wanted some feedback on the resizing at this point.

what algorithm is used to resize? hardware? its better than spline\lanczos\blackman? As I clearly say, it's implemented on the GPU. It is bilinear. I can't answer your last question without knowing your definition of "better".

[Varies]

Be more specific in your concerns.
for example, i want to crop 1 or 3 px ^ ^ of course if i used crop() i must to crop 2 or 4 px, its wrong way ... therefore i used resize(1280,720,0,1,-3,0).
now i resize DGSource("file.dgm",deinterlace=2,resize_w=1280,resize_h=720) and than resize(1280,720,0,1,-3,0) ...again using main cpu.
I can't answer your last question without knowing your definition of "better"
better for upscale or downscale, halo oreols or artifacts when resizing :)

and why i see this picture http://s08.radikal.ru/i181/0908/ed/496ad1f7674c.png last few days ? i don't use other gpu processes...

[Guest]

Try killing the CUVID server.

[Varies]

I restart PC but its dosn't help.

[Guest]

Please post a sample of your source and give me the Avisynth script.

I assume this happens when you load the script into something. What is that something?

[GZZ]

Tested the resize a little and it sure gives med a little speed boost. 1Pass before: ~38.5 fps, now its up to ~42 fps (CPU usage is around 40-45%). On 2Pass before ~15 fps, now it up to ~17 fps. My Geforce is a Passiv cooled 9500 GT (1 Gb ram, DDR2-800). The resize quality looks pretty good, not sure if there is any difference compared to the Spline36Resize(1280,720).Sharpen(0.2) I used before.

My Rig:
Core2Quad: Q9450, OC@3.0 Ghz
GPU: Geforce 9500 GT, 1 GB ram.

I'm happy and keep up the good work.

[Guest]

I've got the source cropping implemented but there's an irritating bug that occurs only in the non-debug build. Why is it never the other way around? :p

Investigating...

[halsboss]

Thankyou for this marvelous tool, neuron2. May I ask if the GPU does deblocking (eg of interlaced sources) and if so whether you are considering it as a new feature at some point ?

Also, am I right in understanding that it only resizes non-interlaced sources or sources where deinterlacing is specified; ie it does not (yet?) resize interlaced sources ?

[Revgen]

Resizing works. Though it's slightly slower than avisynth's bilinear resizer.

[Guest]

Thankyou for this marvelous tool, neuron2. May I ask if the GPU does deblocking (eg of interlaced sources) and if so whether you are considering it as a new feature at some point ? It's not available in the API.

Also, am I right in understanding that it only resizes non-interlaced sources or sources where deinterlacing is specified; ie it does not (yet?) resize interlaced sources ? You could do a double rate deinterlace and resize and then reinterlace it in your script.

[Guest]

Resizing works. Though it's slightly slower than avisynth's bilinear resizer. Are you sure about that? How are you testing that? Nvidia told me it is essentially free. If what you say is true, then there is no point to having this feature. I'll make some tests today.

[Revgen]

Are you sure about that? How are you testing that? Nvidia told me it is essentially free. If what you say is true, then there is no point to having this feature. I'll make some tests today.

Well Vdub said it was faster.

I downsized the KingKong.mkv (VC1) to 1280x720p from 1920x1080p. It was 60fps with Nvidia and 62fps with avisynths bilinear resizer. Then again, I'm using a less expensive 8400 GS and my PC is a quad-core QX6850 with a 1300mhz bus. I only bought the card for cheap just to use your program since my other one was a 6200.

I also tried mpeg-2 clips. Sometimes avisynth and nvidia were the same or avisynth was at 94fps and Nvidia at 91fps.

Difference isn't that big.

[Guest]

You didn't tell me how you measured the fps!

I did a test by making the script and adding assumefps(2000) at the end and then playing it in VirtualDub and measuring the time taken to play it completely. For the one test I did on an 8500 I got this:

GPU resize: 53 seconds
bilinearresize(): 82 seconds

[Revgen]

^Like I said before, I have a very fast CPU combined with a low-end card. It may just be my circumstances. I used virtualdub.

[Guest]

OK, for the third time, I ask:

How are you measuring the fps?

[Revgen]

^Virtualdub.

[Guest]

Did you put the big assumefps at the end? May I see the two scripts?

[Revgen]

Okay, just tried the assumefps thing.

Nvidia was 55fps 1 min 20 secs
Avisynth was 63fps. 1 min 10 secs

Basically, it's even worse when I do that.

Keep in the mind the difference between your card and mine

Yours: http://www.nvidia.com/object/geforce_8500.html
Miine: http://www.nvidia.com/object/geforce_8400.html

The memory bandwith is about half what yours has.

NVidia script

dgsource("King Kong.dgv",resize_w=1280,resize_h=720)
assumefps(2000)

Avisynth script

dgsource("King Kong.dgv")
bilinearresize(1280,720)
assumefps(2000)

[Guest]

Nvidia was 55fps 1 min 10 secs
Avisynth was 63fps. 1 min 20 secs
Avisynth took 10 seconds longer. Isn't that slower?

[GZZ]

Dont know if this is a bug or not (you tell me). If I create a *.dga file, load it up in my avs script. just 2 lines (loadplugin and dgsource) and then start a encoding in x264 (2pass mode). When its encoding (doing 1 pass) I start the avs file in medieplayer classic the image will shutter alot (properly because the gpu is using all its power), but when I close medieplayer classic, my encoding will hang (x264 process goes to 0% cpu) and its not possible to restart the encoding because the CUVIDServer just freeze. DONT try to exit it using the right click on tray icon (you will have to reboot your computer to make it work again, kill it in task manager). Can anyone reproduce all of this ?

[Guest]

You cannot invoke multiple instances like that. It is not supported.

[GZZ]

I just found it, accessing any other AVS file that uses DGDecodeNV.dll doing encoding will kill the CUVIDServer, will multiple instance be allowed or maybe something that will prevent it from freezing, if more then one instance try to access it ?

[Guest]

I could possibly throw an error if the server is busy.

[GZZ]

That would be great, better to get a "Cuda server is busy at the moment" instead of killing an ongoing encoding, just because you execute another avs file that uses DGDecodeNV.

[halsboss]

(re deblocking) It's not available in the API.Darn. Oh well, thanks.

You could do a double rate deinterlace and resize and then reinterlace it in your script.OK ! Do you have a hint on avisynth code to reinterlace the resulting TFF PAL 576p50 after the double rate deinterlace and resize ?

edit1: already asked and answered here http://forum.doom9.org/showthread.php?p=1110741#post1110741

AssumeTFF().SeparateFields().SelectEvery(4, 0, 3).Weave() '-> TFF
AssumeBFF().SeparateFields().SelectEvery(4, 0, 3).Weave() '-> BFF
AssumeTFF().SeparateFields().SelectEvery(4, 1, 2).Weave() '-> BFF
AssumeBFF().SeparateFields().SelectEvery(4, 1, 2).Weave() '-> TFF

edit2: oops, you posted an answer but I only saw it after I posted edit1. Thanks.

[Guest]

OMG, that's been discussed thousands of times:

separatefields()
selectevery(4,0,3)
weave()