DG NV tools

[Blue_MiSfit]

Awesome work, neuron2! This is a very solid version of DGNV!

Here's a question for you - say I need to run a lot of 1080p transcodes at once, say enough to overrun even a 2GB card.

If I install multiple cards in the system, will DGNV spawn instances on all the cards, or is it restricted to one?

Derek

[RedDwarf1]

I just got done saying I know it makes a difference. Are you doubting that or just missed my statement?

I don't doubt that, I wondered how much of a difference it makes.

From what I have seen, memory bandwidth doesn't seem a big issue on my card so far as mine has never gone over 20% with a default 54.4GB/s bandwidth on a 128 bit bus. If the VP cores run at similar speeds where is the difference coming from. Are new VP cores more efficient at decoding and hence faster.

Data would reveal a lot and remove a lot of guesswork on which card to choose.

[Guest]

Here's a question for you - say I need to run a lot of 1080p transcodes at once, say enough to overrun even a 2GB card.

If I install multiple cards in the system, will DGNV spawn instances on all the cards, or is it restricted to one?
Multiple instances are supported of course. But currently a new instance looks at only the first CUDA-enabled card. tormento has already asked for a way to specify which of multiple cards should be used. Now you bring up another mode that should be supported: dynamically find a card that has enough memory for another instance. Adding to the to-do list.

[Sharktooth]

Something i might be interested into.
9500GT comes with stock frequencies at 550/1400/400 (gpu/shaders/memory), while this is pretty "low" i can safely overclock it (without volt mod) to 700/1820/510.
Does the overclocking influence also video decoding speed or has it a dedicated unit for video decoding that's completely independent from core/shaders/memory clocks?

[adiabatic]

Main PC:

[Display]
Processor: Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz (2673 MHz)
Operating System: Windows 7 Professional, 64-bit
DirectX version: 11.0
GPU processor: GeForce GTX 275
Driver version: 257.21
CUDA Cores: 240
Core clock: 648 MHz
Shader clock: 1458 MHz
Memory clock: 1188 MHz (2376 MHz data rate)
Memory interface: 448-bit
Total available graphics memory: 3707 MB
Dedicated video memory: 896 MB GDDR3
System video memory: 0 MB
Shared system memory: 2811 MB
Video BIOS version: 62.00.60.00.70
IRQ: 24
Bus: PCI Express x16 Gen2

Secondary PC (my old workhorse)
[Display]
Processor: Intel(R) Core(TM)2 CPU 4400 @ 2.00GHz (1995 MHz)
Operating System: Windows Vista (TM) Home Premium, 32-bit (Service Pack 2)
DirectX version: 10.1
GPU processor: GeForce 8600 GT
Driver version: 257.21
CUDA Cores: 32
Core clock: 540 MHz
Shader clock: 1188 MHz
Memory clock: 700 MHz (1400 MHz data rate)
Memory interface: 128-bit
Total available graphics memory: 1021 MB
Dedicated video memory: 256 MB GDDR3
System video memory: 0 MB
Shared system memory: 765 MB
Video BIOS version: 60.84.51.00.00
IRQ: 16
Bus: PCI Express x16

[MrVideo]

Here is mine, for the new quadcore. The same graphics card is in the dual core system.

[Display]
Processor: AMD Phenom(tm) II X4 965 Processor (3411 MHz)
Operating System: Microsoft Windows XP, 32-bit (Service Pack 2)
DirectX version: 9.0
GPU processor: GeForce GT 240
Driver version: 257.21
CUDA Cores: 96
Core clock: 550 MHz
Shader clock: 1340 MHz
Memory clock: 1700 MHz (3400 MHz data rate)
Memory interface: 128-bit
Memory: 512 MB
Memory type: GDDR5
Video BIOS version: 70.15.2C.00.51
IRQ: 24
Bus: PCI Express x16 Gen2

[Guest]

Memory interface: 448-bit
Hit man is on the way. :)

I think we should hold off further reports and do what RedDwarf1 suggested. I will get a sample clip prepared with instructions for playing it in DGIndexNV and then we will report the data above together with our FPS. Stand by...

[Guest]

Does the overclocking influence also video decoding speed or has it a dedicated unit for video decoding that's completely independent from core/shaders/memory clocks? Good question. I'll see what Nvidia has to say about it. I'll also ask if there is a way to determine the VP engine clock rate.

[Audionut]

I asked the rivatuner (http://www.guru3d.com/category/rivatuner/) author about these things not long after dgindexnv was released and got this response.

Hmm, interesting. I did some researches tonight and it looks like there is indeed separate 4xx MHz clocked domain in the chip which can be used to feed VP independently of SPs. However, forget about controlling it: drivers doesn't provide overclocking support for it and implementing low-level way doesn't worth the efforts from my POV.

http://forums.guru3d.com/showthread.php?t=274311

Perhaps a few of you guys should register there and bug the dev about adding overclock support for the chip.

[adiabatic]

Hit man is on the way. :)


It is a nice card :) It rips through SETI work units like crazy too!

[tormento]

My home generic use PC:

[Schermo]
Processore: Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz (3399 MHz)
Sistema operativo: Windows 7 Ultimate, 64-bit
Versione DirectX: 11.0
Processore GPU: GeForce GTX 275
Versione del driver: 257.21
Nuclei CUDA: 240
Clock principale: 666 MHz
Clock ombreggiatore: 1476 MHz
Clock memoria: 1161 MHz (frequenza dei dati 2322 MHz)
Interfaccia memoria: 448 bit
Memoria grafica totale disponibile: 4095 MB
Memoria video dedicata: 896 MB GDDR3
Memoria video di sistema: 0 MB
Memoria di sistema condivisa: 3199 MB
Versione BIOS video: 62.00.60.00.01
IRQ: 16
Bus: PCI Express x16 Gen2

My office PC:
[Schermo]
Processore: Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz (3601 MHz)
Sistema operativo: Windows 7 Ultimate, 64-bit
Versione DirectX: 11.0
Processore GPU: Quadro FX 3700
Versione del driver: 258.96
Nuclei CUDA: 112
Clock principale: 675 MHz
Clock ombreggiatore: 1600 MHz
Clock memoria: 975 MHz (frequenza dei dati 1950 MHz)
Interfaccia memoria: 256 bit
Memoria grafica totale disponibile: 3323 MB
Memoria video dedicata: 512 MB GDDR3
Memoria video di sistema: 0 MB
Memoria di sistema condivisa: 2811 MB
Versione BIOS video: 62.92.16.00.04
IRQ: 24
Bus: PCI Express x16 Gen2
I think I should kill the NVidia italian translator.

P.S: 258.96 works ok.

[lych_necross]

My Specs:

[Display]
Processor: Intel(R) Core(TM)2 CPU 6400 @ 2.13GHz (2133 MHz)
Operating System: Windows 7 Ultimate, 64-bit
DirectX version: 11.0
GPU processor: GeForce 9800 GTX/9800 GTX+
Driver version: 258.96
CUDA Cores: 128
Core clock: 760 MHz
Shader clock: 1890 MHz
Memory clock: 1125 MHz (2250 MHz data rate)
Memory interface: 256-bit
Total available graphics memory: 2303 MB
Dedicated video memory: 512 MB GDDR3
System video memory: 0 MB
Shared system memory: 1791 MB
Video BIOS version: 62.92.62.00.61
IRQ: 16
Bus: PCI Express x16

I might upgrade my video card soon. (I'm thinking about replacing it with the GeForce GTX 460 1GB)

[tormento]

@neuron2

Just a crazy idea. Do you think you could implement (by your internal nvidia contacts) some compatibility between DGNV and VReveal?

Nvidia is offering a free SD version and the HD is paying available.

The destination should be feeding AviSynth with stabilized video from VReveal... A dream...

[Clumpco]

Hi Neuron 2,
Just been reading all the changes and hard work that you have been doing while I have been sunning myself on holiday. Thanks!

Using the x86 version under Windows7 I see the 'slowdown' if I leave the windows visible, if I minimise I get a 2x speed gain. Thanks for the tip/workaround.

I wonder if it isn't your progress bar that slows things down, it seems to flash a lot while indexing (indicating that the control is being redrawn rather often?) - have you tried disabling that but leaving the frames/time elapsed/remaining active?

Cheers

[Guest]

Using the x86 version under Windows7 I see the 'slowdown' if I leave the windows visible, if I minimise I get a 2x speed gain. Thanks for the tip/workaround.

I wonder if it isn't your progress bar that slows things down, it seems to flash a lot while indexing (indicating that the control is being redrawn rather often?) - have you tried disabling that but leaving the frames/time elapsed/remaining active? Thanks for the information and suggestion. I will look into it as I now have a Win7 system.

[stax76]

The latest build still crashes arbitrary opening/closing scripts in fast sequence, maybe it's just my system (GT 9400 card rather new driver). The server thing was working good. There is a new StaxRip build with some changes, removed DGSource, added DGMultiSource etc., I hope you can fix it.

http://forum.doom9.org/showthread.php?p=1417164#post1417164

[Guest]

Please give me explicit instructions for making the problem happen.

[stax76]

Well, maybe I found something, instead of only querying frame count etc. actually querying a image seem to help.

[stax76]

That seem to help indeed, not a single crash yet and before it was almost always crashing, yesterday only sometimes. It must be some initialization problem.

[Guest]

>Well, maybe I found something, instead of only querying frame count etc.
>actually querying a image seem to help.

I don't understand what you are talking about here.

I'll wait then for you to tell me whether to proceed on a bug hunt.

I can hold down the F2 key in VirtualDub. :)

[stax76]

Solved I would say, it was as simple as just querying a image. :)

Dim a As New AVIFile
a.Open(Path)
a.GetBitmap()
Framerate = a.FrameRate
Frames = a.FrameCount
FourCC = a.FourCC
Size = a.FrameSize
ErrorMessage = a.ErrorMessage
a.Close()

I can hold down the F2 key in VirtualDub.

Because VD always queries a image frame.

[Turtleggjp]

>Well, maybe I found something, instead of only querying frame count etc.
>actually querying a image seem to help.

I don't understand what you are talking about here.

I'll wait then for you to tell me whether to proceed on a bug hunt.

I can hold down the F2 key in VirtualDub. :)

I think VirtualDub always loads the first frame immediately, which is why Stax76's loading of the first frame works better for him.

[Guest]

Now that DGDecodeNV.dll and the CUVID Server are decomposing gracefully in their new earthy abode, will anyone go postal if I rename DGMultiSource() to just DGSource()?

Maybe I can make then synonymous so both are accepted.

[Blue_MiSfit]

Good idea. I think synonymous is the way to go. I've been through that particular headache in the past with other functions :)

[lych_necross]

Now that DGDecodeNV.dll and the CUVID Server are decomposing gracefully in their new earthy abode, will anyone go postal if I rename DGMultiSource() to just DGSource()?

Maybe I can make then synonymous so both are accepted.
I don't have a problem with that as long as you could make them synonymous.

[mikeyakame]

Here's a capture from GPU-Z, has some more detailed info.

http://gpuz.techpowerup.com/10/07/14/43b.png

I actually had to clock my memory down to 1250mhz, because 1300mhz was giving occasional decoder crashes over long encodes, like say a 6-8hr x264 encode.

And heres a sensor screen shot while doing a source/resize from an indexed mkv (1280x720 x264 ~2mbps bitrate) that I encoded.

http://sadpanda.us/images/168013-QXENWVR.gif

This is basically the contents of the avs and running a video analysis pass in virtualdub for a filter output.


SetMTMode(0,4)
DGMultiSource("G:\temp\encoding\angel\dgi\ep01_mkv.dgi",resize_w=704,resize_h=400)
...


And here's a load shot using 64-bit Virtual Dub doing a video analysis pass. The indexed source is a 1080p24 M2TS from a japanese anime blu-ray that I own. If I recall average bitrate was around the 37-38mbps mark. I'm not too sure if this is a factor on video engine load though? Maybe you might know the answer to that. I know higher bitrate results in higher decode load, but not sure if it applies for frameserving or not.
Video Memory use for this 1080p source was ~230mb. It jumped from 210mb -> ~443mb.

http://gpuz.techpowerup.com/10/07/14/dz4.png

And one final one. Vdub 64-bit + same source, but using 4 multisource inputs in the avs.
Vram usage is 210mb -> 583mb, so ~ 370mb for 4 dgi source inputs at 1920x1080. Just in case you ask "Why 4 sources!"


DGMultiSource("G:\temp\encoding\angel\dgi\bd\ep01-02.dgi").Trim(37488, 72767)
pt2 = last.Trim(2160, 32759)
pt1 = DGMultiSource("G:\temp\encoding\angel\dgi\bd\ep02_cdop.dgi").Trim(24, 2183)
pt3 = DGMultiSource("G:\temp\encoding\angel\dgi\bd\ep02_cded.dgi").Trim(24, 2183)
pt4 = DGMultiSource("G:\temp\encoding\angel\dgi\bd\ep02_web_pv.dgi").Trim(24, 383)
pt1++pt2++pt3++pt4


Basically i've just trimmed out sections and spliced in bonus feature replacements (which are extras on the blu-ray and are same frame length as original ones). Since they need to be frame exact and I needed to output merged wavs, this is easier to work with than appending multiple m2ts files into 1 index.

http://gpuz.techpowerup.com/10/07/14/644.png

Thought it might be interesting for you to see, Neuron.

[Guest]

Thanks, it is indeed interesting.

[Guest]

Maybe I can make then synonymous so both are accepted. It turned out to be trivial to make them synonymous.

Next question: Will anyone go ballistic if I rename the DLL?

[rack04]

Next question: Will anyone go ballistic if I rename the DLL?

Nope.

[Audionut]

Self reminder to delete old dll's from plugin dir when Don renames.

[cweb]

It turned out to be trivial to make them synonymous.

Next question: Will anyone go ballistic if I rename the DLL?

Not me.. you can rename it.

[Guest]

Everything is renamed now for 2021 and DGMultiSource is retained as a synonym for DGSource.

[laserfan]

Everything is renamed now for 2021 and DGMultiSource is retained as a synonym for DGDecode.I guess this means my batch files/cmds won't need to be changed, but I'm gonna do it anyway just to avoid any confusion (and I'm easily confused ;)) in the future.

Thanks DG, looking forward to the new version.:)

[Blue_MiSfit]

Awesome :) This utility is getting some fantastic polish development over the last few weeks!

I'm about to buy a GTX 460 as well for my home machine, so it should prove interesting :devil:

Derek

[MOS-Marauder]

My Specs:

[Anzeige]
Prozessor: Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz (3404 MHz)
Betriebssystem: Windows 7 Ultimate, 64-bit
DirectX-Version: 11.0
GPU-Prozessor: GeForce GTX 260
Treiberversion: 257.21
CUDA-Kerne: 192
Kerntakt: 576 MHz
Shadertakt: 1242 MHz
Speichertaktfrequenz: 999 MHz (1998 MHz Datenrate)
Speicherschnittstelle: 448-Bit
Gesamter verfügbarer Grafikspeicher: 3711 MB
Dedizierter Videospeicher: 896 MB GDDR3
System-Videospeicher: 0 MB
Freigegebener Systemspeicher: 2815 MB
Video-BIOS-Version: 62.00.61.00.00
IRQ: 16
Bus: PCI Express x16

Chris

[Alf Bundy]

Hi,

Is there a limit in the resolution supported by Nvidia GPUs ?

I made a x264 encode from the 500 frames of "Crowd Run" (3840x2160 @ 24Mbps).
The video is playable using MediaPlayerClassic-HomeCinema's internal decoder (but only with the standard decoder, not the DXVA one).

And loading the elementary stream into DGNV results in this :

http://img121.imageshack.us/img121/3055/error400.png


I wanted to compare the decoding speed between my little 9400GT and my 9800GTX+ using a monstrous video (to make the differences more clear).

Maybe your contacts at Nvidia could tell us what's the maximum resolution allowed for GPU decoding ?

:thanks:

[Blue_MiSfit]

Yes, as far as I know you can only decode up to 1920x1080 using the nVidia VP, although bitrates and references up to Level 5+ are allowed, if I recall correctly... don't quote me on that.

Long story short, I wouldn't expect 4k to decode properly :)

Derek

[Guest]

Small correction: 1920 x 1088.

[tormento]

Would a frame split and interleave work for 4k resolution?

[stax76]

There is a new StaxRip build finally supporting DGMultiSource:

http://forum.doom9.org/showthread.php?p=1417817#post1417817

[Guest]

Would a frame split and interleave work for 4k resolution? I don't know what you are talking about, so I suppose the answer is no.

[Groucho2004]

Got my Asus GT 240 GDDR5 (512MB) and Don sent me a license yesterday. Everything is working as it should - nice work Don.

One question - How do you guys go about removing 3:2 pulldown? I have been using decomb with "AssumeT(B)FF().Telecide().Decimate()" so far and today I compared it to "fieldop=1" (force film) in DGMultiDecodeNV.dll. Motion seems a bit more fluid with decomb but maybe I'm imagining things. What are your experiences?

[cweb]

Got my Asus GT 240 GDDR5 (512MB) and Don sent me a license yesterday. Everything is working as it should - nice work Don.

One question - How do you guys go about removing 3:2 pulldown? I have been using decomb with "AssumeT(B)FF().Telecide().Decimate()" so far and today I compared it to "fieldop=1" (force film) in DGMultiDecodeNV.dll. Motion seems a bit more fluid with decomb but maybe I'm imagining things. What are your experiences?


Here's what I did for NTSC (video not film if I recall well) using tfm (http://avisynth.org.ru/docs/english/externalfilters/tivtc_tfm.htm) instead.

dgmultisource("title01.dgi",deinterlace=0)
AssumeTFF()
tfm(order=1)
tdecimate()

It worked better for me than telecide/decimate. YMMV of course. Best thing is to try both methods for a small clip and see what works best for your material.

[Guest]

One question - How do you guys go about removing 3:2 pulldown? I have been using decomb with "AssumeT(B)FF().Telecide().Decimate()" so far and today I compared it to "fieldop=1" (force film) in DGMultiDecodeNV.dll. Motion seems a bit more fluid with decomb but maybe I'm imagining things. What are your experiences? It all depends on whether you have soft or hard 3:2 pulldown and whether it is regular or irregular. This is really not a topic for this thread and has been extensively discussed elsewhere. You can start a new post in Avisynth Usage about it. Be sure to post an unprocessed source sample with a high motion scene.

[sumawo13]

When I try to open anything literally nothing happens.

I'm using Windows 7 Ultimate 64-Bit with an 8400GS.

[Guest]

Read the release notes.

[hydra3333]

I upgraded to 2020 from 2018, and can open a PAL 1440x1088i .mpg, but when I try to save a project, nothing happens. Previous versions worked fine.

Latest nvidia, plain standard vanilla XP-sp3 all latest updates, re-did and checked the licenses and the show the same long numbers.

So,
1. extracted v2020 into a folder with the licenses and opened the .mpg - save a project and it just sits there.
2. over-wrote v2020 with v2019 and did the same - same result.
3. over-wrote v2019 with v2018 and did the same - it worked fine.

It did ask a question about cropping to which I clicked no. Tried again with yes not no change in result.

Oh, if I run 2020 and try to close it's window, a couple of times , it then hangs.

Fully repeatable on 2nd licensed machine with latest nvidia, plain standard vanilla XP-sp3 all latest updates. Didn't spot anything obvious in the release notes.
q9450, 8800GT
q6600, 8600GT

It seems that the version 2019/2020 is broken somehow. Suggestions ?

[Guest]

I need a stream to duplicate your issue. Please be sure that you can duplicate the issue with the stream that you provide to me. Thank you.

[hydra3333]

OK I will try to extract it somehow and post it somewhere (any preference?)

[Guest]

Cut it with DGSplit and upload it to mediafire.com.

As I said, test it with the split file so I don't waste time on a stream that doesn't fail. Thank you.