Ultimate PC spec for VirtualDub?

[discmeister]

Hi all,

I've seen various discussions here about which operating system to use for VirtualDub. But while the base PC specifications are reasonably well known, I was wondering if there is an optimum PC spec for running it? I only ask because have an i7-7700k with 32Gb RAM that struggles to get beyond 8fps when it's conducting a reasonably standard set of AVISynth instructions in VirtualDub and I'm wondering if there's any way to get that figure up. Multi-CPU motherboard, for example?

Kind regards to all,

Discy

[lordsmurf]

For filtering, yes? It depends on what's being run. Details needed.

[discmeister]

Well, as an example, the avisynth file provided to me several months back by sanlyn.

Quote:

### --- Remove and configure new borders ----#
Crop(16,2,-14,-6).AddBorders(16,4,14,4)

### --- contrast, gamma, hue, saturation issues ----#
ColorYUV(cont_y=-20,off_y=-8) Levels(20, 1.2, 255, 16, 250, dither=true, coring=false) Tweak(Sat=0.90, starthue=165, endhue=180, dither=true, coring=false)
ConvertToYV24(interlaced=true) SmoothTweak(hue1=5)

### --- deinterlace setup and noise reduction ----#
ConvertToYV12(interlaced=true) AssumeTFF() QTGMC(preset="medium",EZDenoise=8,denoiser="dfttes t",ChromaMotion=true,border=true,\ ChromaNoise=true,DenoiseMC=true,GrainRestore=0.3,S tabilizeNoise=true)

### --- smooth excessive combing ----#
Vinverse2()

### --- smooth aliasing and jaggies ----#
ConvertToYUY2(interlaced=false) Santiag(2,2)

### --- add film-like fine grain to mask over-filtering ----#
AddGrainC(1.5,1.5)

###---Remove borders, resize to fill frame ---###
Crop(16,4,-14,-4) Spline36Resize(704,576)

###---To RGB for VirtualDub filters ---###
ConvertToRGB32(interlaced=false,matrix="Rec601")

Did anyone have any thoughts on what sort of spec would make the adjustments above go more quickly than 8fps? Or is that, in fact, a decent figure?

I'm wondering whether picking up an old server CPU - a Xeon with lots of cores and threads - secondhand might be a better route than the modern i7 that I have in my PC.

Grateful for any advice - I have plenty to convert, so any gains in this area will save a huge timesoak in PC processing...

Discy

[JPMedia]

If you want a processor that is natively compatible with Windows 7, you should look into the 4 core Intel i7 6700k or the 6 core Intel i7 7800x.

More recent processors require modifying the Windows 7 installer ISO to have driver compatibility.

Check out this VideoHelp thread for more info.

[themaster1]

I don't know what you use to capture (usb,pci) but it's well known W7 drivers are subpar (blown highlights guaranted), Xp recommanded.
Virtualdub don't need a horse for reliable caps (i'm on an old phenom2 X4 945) it needs a stable signal (tbc'd) and 2 hdd's at a minimum (1 for the Os one for the caps) to avoid dropped frames

[JPMedia]

Quote:

Originally Posted by themaster1

I don't know what you use to capture (usb,pci) but it's well known W7 drivers are subpar (blown highlights guaranted), Xp recommanded.
Virtualdub don't need a horse for reliable caps (i'm on an old phenom2 X4 945) it needs a stable signal (tbc'd) and 2 hdd's at a minimum (1 for the Os one for the caps) to avoid dropped frames

I built an XP capture rig.

You want Windows 7 for VirtualDub and AviSynth processing post capture.

Also not sure what you mean about with these blown highlights.

When you say "Ultimate PC spec for VirtualDub," are you talking about capture or post processing?

[discmeister]

Thanks for your responses. I have a sacrificial laptop that I use for capture and it works fine. It captures to Lagarith lossless compression.

No, I'm asking what the ultimate spec of PC would be for VirtualDub to load an AVISynth file with the filters as outlined earlier in this thread and then reoutput the retouched file, again with Lagarith lossless compression (although in RGB, as per the filters).

I'm currently getting about 8fps. I was just wondering whether this is pretty decent or whether, say, building a multi-CPU Xeon (or even a single-CPU Xeon with extra cores/threads) system would help.

Thanks again for all advice.

Kind regards,

Discy

[Feedbucket]

That's normal. You're witnessing QTGMC doing what it does - dialing back the noise reduction settings might speed things up, but you'll lose whatever visual effects those settings cause.

AviSynth isn't multithreaded straight out of the box, so throwing more cores at it probably won't do much more than allow you to multitask. If you've got the patience to figure out how to update all of QTGMC's required plugins to 64-bit, AviSynth+ might give you an improvement - though I've found that the specific cocktail of plugins that QTGMC uses does not directly benefit from multithreading as much as some single atomic functions do, and at times it will produce some rendering errors (chain of execution gets jumbled, I think). Raw clock speed is for the most part the better way to go if its an option.

Edit to above: I just ran your script and it looks like it's the double-whammy of QTGMC and Santiag (both use NNEDI3, which is sloooow). I did use AviSynth+ in both single and multi-threaded modes and multi-threaded appears to run maybe 20% faster.

[Angies_Husband]

To reiterate what others have said... the low FPS is normal because most things run single threaded.

I have a Dual Xeon X5675 HP workstation (12 cores total, 24 threads) and it does NOT help for running normal Avisynth stuff. It is great for stuff that can take advantage of the multiple cores though...

[themaster1]

Quote:

Originally Posted by Angies_Husband

To reiterate what others have said... the low FPS is normal because most things run single threaded.

I have a Dual Xeon X5675 HP workstation (12 cores total, 24 threads) and it does NOT help for running normal Avisynth stuff. It is great for stuff that can take advantage of the multiple cores though...

Yeah and i might add dealing with HD 1080p (or worse interlaced) you're in it for 10 hours or more for 30min of video. When you love you don't count they say...but i do

I read avisynth+ gives a great boost (memory management etc..) but the backwards compatibility of plugins is the eternal problem.


I can't wait the AI networks (available to everybody not just researchers) take over images/video processing example:

https://www.dpreview.com/videos/1559702271/researchers-use-ai-to-brighten-ultra-low-light-images-without-adding-noise