How to Speed Up McTemporalDenoise?

[VideoFanatic]

I have an 8 core PC. I'm using Avisynth MT. For the 1st line and number I tried all methods below and above 5 and they were slower. For the 2nd number I tried all numbers above and below 16 and they were slower. Is the 3rd line correct or should I modify it?

setmtmode(5,16)
Mpeg2Source("J:\2 = New\Raw August 06 2001.d2v", CPU=6)
setmtmode(2,0)
McTemporalDenoise(settings="medium", interlaced=true)
ConvertToYUY2(interlaced=true)
TemporalSoften(4,4,8,15,2)
ConvertToYV12(interlaced=true)
AssumeTFF() # or AssumeBFF() depending on the source

# Deinterlace
Load_Stdcall_plugin("C:\Program Files (x86)\AviSynth 2.5\plugins\yadif.dll")
QTGMC(Preset="Ultra Fast") 
Vinverse()

#Re-interlace
SeparateFields()
SelectEvery(4,0,3)
Weave()

RemoveSpots()
LSFMod(strength=50)
AddGrainC(var=1.0,uvar=1.0)

Assuming my code is correct, can anyone please suggest a way to speed up McTemporalDenoise? Here are some speeds I'm getting:

• McTemporalDenoise on it's own: 10.40 FPS
• Everything except McTemporalDenoise: 35 FPS
• Everything & Super Fast for QTGMC (strange how it's quicker than Ultra Fast): 5.50 FPS
  

So it's McTemporalDenoise that's slowing everything else down.

I'm currently getting 10.40 FPS and it's taking around 5 hours 40 minutes for a 1 hour 40 minute video. I would like to get this down to 4 hours 30 minutes otherwise I can only do 4 videos in 24 hours instead of 5 which will slow my work down.

I'm open to any suggestions. I've got a CUDA VP5 graphics card and I've just bought DGDecNV and I'm waiting on my licence. Do you think that will give a faster FPS than the free version - DGMPGDec?

I think there's a way to enable the GPU for McTemporalDenoise but I don't know how to install it to get it working.

I use Simple X264 Launcher to encode my videos and it just does 1 pass.

[poisondeathray]

I've got a CUDA VP5 graphics card and I've just bought DGDecNV and I'm waiting on my licence. Do you think that will give a faster FPS than the free version - DGMPGDec?

It won' t make much of a difference in your case - the bottleneck isn't decoding, and the CPU required to decode a SD MPG source is <1%

I think there's a way to enable the GPU for McTemporalDenoise but I don't know how to install it to get it working.

GPU=true . It only speeds up the fft3dgpu part (not a huge speed increase and it can cause weird artifacts so be careful - fft3dgpu doesn't always stable results in my experience)


If some filters are bottlenecks and you have slack CPU usage, you can do parallel processing (encode a few instances or different episodes simultaneously)

[VideoFanatic]

Could you please tell me how to install fft3dgpu? I ran the fft3dgpu0.8.2a.exe installer but it still doesn't work.

[poisondeathray]

Could you please tell me how to install fft3dgpu?

use google!

fft3dgpu
http://forum.doom9.org/showthread.php?t=89941


To activate it in MCTD use gpu=true

The work normally done by "vanilla" fft3dfilter is partially sped up by fft3dgpu instead . It's only a fraction of what goes in the MCTD , so it might only be a few % faster

[VideoFanatic]

I was already aware of those pages but I ran the fft3dgpu0.8.2a.exe installer but it still doesn't work. When I try to encode a video it gives the following error: PROCESS EXITED WITH ERROR CODE: -1073741819


So I thought you might know a different way to install it to get it working.

[mammo1789]

just put FFT3dGPU.dll and fft3dGPU.hlsl ( dont forget this one ) in avisynth plugin folder, you don't have to run the installer you can just download those 2 http://forum.doom9.org/showthread.php?t=89941
then just use MCTemporalDenoise(settings="high", gpu=true)

[VideoFanatic]

I already did that before and I got the error message as mentioned before.

[poisondeathray]

Do you have directx installed ?

[VideoFanatic]

I have DirectX 11 installed. I see the following files in SysWOW64:

d3dx9_30.dll
d3dx9_31.dll
D3DX9_42.dll

[mammo1789]

I'm currently getting 10.40 FPS and it's taking around 5 hours 40 minutes for a 1 hour 40 minute video. I would like to get this down to 4 hours 30 minutes otherwise I can only do 4 videos in 24 hours instead of 5 which will slow my work down.

i don't think that you can get faster than that i get around 3,5 fps on "high" with quad core 3,4 ghz if anyone else knows how to speed it up more please say so.

[mammo1789]

Do you have directx installed ?

i don't think that direct x is the problem he has windows 7 I guess so it is installed. it doesn't hurt to try it to install the latest version http://www.microsoft.com/en-us/download/details-b.aspx?id=35 , maybe it is the card driver or the card it self what graphic card do you have?

[VideoFanatic]

If some filters are bottlenecks and you have slack CPU usage, you can do parallel processing (encode a few instances or different episodes simultaneously)

Are you referring to just encoding 2 videos at the same time? I did that and all it did was half the FPS for each video saving no time.

I have the following graphics card: Gigabyte GV N520SL-1GI

[poisondeathray]

If some filters are bottlenecks and you have slack CPU usage, you can do parallel processing (encode a few instances or different episodes simultaneously)

Are you referring to just encoding 2 videos at the same time? I did that and all it did was half the FPS for each video saving no time.

Basically yes

If 1 instance uses less than 100% (open up taskmanager as you are encoding) , then you can benefit from parallel encodes. For example if it only uses ~50% for 1 instance, then you should be able to approximately double the speed

There can be bottlenecks anywhere, encoding settings, certain filters . If you can't remove a certain bottleneck, your only option to speed it up is to process in parallel , or faster hardware

[VideoFanatic]

I have an 8-core 3.1 Ghz CPU. The CPU usage is at 100% in Task Manager which suggests the script is using all of the CPU's power. So do you think getting a faster CPU would speed it up?

[poisondeathray]

I have an 8-core 3.1 Ghz CPU. The CPU usage is at 100% in Task Manager which suggests the script is using all of the CPU's power. So do you think getting a faster CPU would speed it up?

Then you have no room. You are using all your resources (under that situation with the current script and encoding settings) . So parallel processing won't help this case

Overclocking it or using a higher clocked CPU model will help - a faster clock speed always helps , but more cores only help if the programming has been optimized for it (many programs and filters are single threaded or poorly multithreaded)

[mammo1789]

I have the following graphics card: Gigabyte GV N520SL-1GI

sorry but that's mediocre card at best memory bus is only 64 bit wide it is low performance card so it is better to not use gpu=true

I have an 8-core 3.1 Ghz CPU. The CPU usage is at 100% in Task Manager which suggests the script is using all of the CPU's power. So do you think getting a faster CPU would speed it up?

i don't think so

[VideoFanatic]

I was told that as long as it's a CUDA card and it has VP5 then that's all you need for DGDecNV. So if it works for that then why wouldn't it work for McTemporalDenoise in the free version - DGMPGDec when enabling gpu=true?

[poisondeathray]

I was told that as long as it's a CUDA card and it has VP5 then that's all you need for DGDecNV. So if it works for that then why wouldn't it work for McTemporalDenoise in the free version - DGMPGDec when enabling gpu=true?

These are completely different things and not comparable

DGNVTools uses VPx engine for decoding (and some optional processing like deinterlacing), MCTD doesn't make use VPx and doesn't make use of GPU for decoding (it has no decoding functionaltiy)

[VideoFanatic]

So is the only reason I'm getting this error message is because my graphics card isn't good enough? I didn't see anything mentioned in McTemporalDenoise that said you required a good graphics card.

Do you think I could just encode my video as normal but then also encode another video at the same time via DGDecNV? That way the 2nd video only uses the Graphics card and won't slow down the encoding speed of the 1st video?

[poisondeathray]

So is the only reason I'm getting this error message is because my graphics card isn't good enough? I didn't see anything mentioned in McTemporalDenoise that said you required a good graphics card.

No idea what the error message indicates, but your GPU should be compatible

Do you think I could just encode my video as normal but then also encode another video at the same time via DGDecNV? That way the 2nd video only uses the Graphics card and won't slow down the encoding speed of the 1st video?

Won't help. You'll just slow down the 1st encode . You're already at 100%. DGNVTools doesn't encode, it decodes. You need CPU to encode. Even if you offload that decoding, remember a SD MPEG2 source takes <1% CPU to decode (fractional , more like <0.5%) . So that's the speed increase you might get at maximum

Situations where you might get speed increase from DGNVTools are from HD files, and using certain encoding settings . In some situations and VPx versions, it's actually slower - the decoding speed becomes the bottleneck for a straight encode (old VP3-4 cards were capped at ~60FPS, where as using software decoding you could get 200-300FPS for SD files decoding speed)

[VideoFanatic]

OK so my card should work with McTemporalDenoise(settings="medium", interlaced=true, GPU=true) . However when I try to encode a video it gives the following error: PROCESS EXITED WITH ERROR CODE: -1073741819.

So I'm still trying to get this to work. Any suggestions? I ran the fft3dgpu0.8.2a.exe installer and I also tried copying the files manually.

[jagabo]

Did you put fftw3.dll in your SysWow64 folder?

[VideoFanatic]

Yes

[mammo1789]

OK so my card should work with McTemporalDenoise(settings="medium", interlaced=true, GPU=true) . However when I try to encode a video it gives the following error: PROCESS EXITED WITH ERROR CODE: -1073741819.

if the question is will it work the answer is yes

The following cards will not work:

Nvidia:
TNT
TNT2
Geforce 256
GeForce2 Ultra, Ti, Pro,MX,Go and GTS
Geforce3 Ti 200, Ti 500
GeForce4 Ti, MX, Go

Ati:
Radeon 7xxx
Radeon 8xxx
Radeon 90xx
Radeon 92xx

Matrox:
G2xx
G4xx
G5xx
maybe Parhelia
The following should work:

Nvidia:
Geforce FX 5xxx
Geforce 6xxx
Geforce 7xxx

Ati:
Radeon 9500
Radeon 9550
Radeon 9600
Radeon 9700
Radeon 9800
Radeon Xxxx
Radeon X1xxx

if the question is it a good card ?( cuda benefit ) then answer is no it has only 48 ( cuda ) cores and has very narrow 64 bit memory controller which makes putting 1 gb on this card ( and gddr3 by the way not gddr5 ) useless. For gpu to take benefit of decoding encoding ( cuda nvidia, stream ati those are just marketing names for the same thing you have to use software plugin that can take benefit of it, otherwise as poi mentioned "software" only cpu will be faster ). And even if that is the case your card is slow and not powerful enough mine has 1440 (1600 ) cores i think that minimum for "cuda" rendering is geforce gtx285 and preferably one in the domain of geforce gtx 570, 580 or 660 and up in the new generations cards to see any real benefits. I also use geforce 285 as secondary card for cuda based programs ( like mpeg2 or h264 encoding )

question? if you put without gpu=true does it work or you get the error anyway

[VideoFanatic]

Yes it works without gpu-true but when I add gpu=true I get the error.

[mammo1789]

Yes it works without gpu-true but when I add gpu=true I get the error.

then maybe as I sad it is the graphic card being not powerful enough i don't know how fft3dgpu really uses the cards cores, but if without it you don't get the error then maybe it is the card.
As far as I can see the error is not filter related

And you have pretty powerful cpu so use that then