QTGMC Multi-Threaded Usage optimization

[maudit]

hi, I use sometimes that command in order to deinterlace some dvds.
QTGMC(Preset="Slow").SelectEven()

I'm currently using it with the nth dvd but this time it seems the cpu does not 100% loads.

I've read the instructions here:
http://avisynth.nl/index.php/QTGMC#Multi-Threaded_Usage
but I didn't understand it very well...

I tried to add the "EdiThreads Y", such us like this way:
QTGMC(Preset="Slow", EdiThreads=8).SelectEven(),
but it seems nothing change so much. tried with the 4 value too, is the same.

So maybe there's another command to use in combination with EdiThreads, in order to optimize QTGMC for multithreading.
Can you please help me?

[VoodooFX]

Use Prefetch, for example for CPU with 4 cores:

QTGMC(Preset="Slow").SelectEven()
Prefetch(3)

[Selur]

Details might be helpful.
What does your complete script look like?
What resolution has your input?
What cpu are you using?
What cpu usage are you seeing?
What Avisynth version are you using?

Higher CPU usage doesn't necessarily mean higher processing speed.
Changing the ediThreads doesn't really do much for me when processing SD content. (anything above 2 even slows down processing, but increases the cpu usage and thread count)
I for example get ~35fps for QTGMC on NTSC 30fps content using Avisynth++ 64bit . (~50% cpu usage)

Did you use AVSMeter you check CPU usage, Thread count, fps etc?
If not, do that. Run it 3 times per setting to get the average usage)

Cu Selur

Ps.: as a side note: using Vapoursynth I get ~42-44fps on the same SD content with QTGMC and higher cpu usage.

[maudit]

Details might be helpful.
What does your complete script look like?
What resolution has your input?
What cpu are you using?
What cpu usage are you seeing?
What Avisynth version are you using?

What does your complete script look like?

complete script:

LoadPlugin("\MeGUI-2913-64\tools\ffms\ffms2.dll")
FFVideoSource("\title_t00_track1_[eng].mkv", fpsnum=25, fpsden=1, threads=1)
QTGMC(Preset="Slow").SelectEven()
crop(0, 74, -8, -74)
FillBorders(0,1,1,0)
bbmod(2,1,2,2,128,8)

What resolution has your input?
It's a PAL 4:3 dvd - 720x576


What cpu are you using?
AMD Ryzen 5 3500U - is a mobile apu 4 core, 8 threads


What cpu usage are you seeing?
Around 70%


What Avisynth version are you using?
it should be this: Avisynth+ 3.7.0


trying to use that AVSMeter3090...

[Selur]

Like VoodooFX suggested try:
You can try:

SetFilterMTMode("DEFAULT_MT_MODE", MT_MULTI_INSTANCE)
LoadPlugin("\MeGUI-2913-64\tools\ffms\ffms2.dll")
FFVideoSource("\title_t00_track1_[eng].mkv", fpsnum=25, fpsden=1)
QTGMC(Preset="Slow").SelectEven()
crop(0, 74, -8, -74)
FillBorders(0,1,1,0)
bbmod(2,1,2,2,128,8)
PreFetch(4)
return last

see: http://avisynth.nl/index.php/AviSynth+#Multithreading_.28MT.29_Notes

Using AVSMeter you can additionally test whether using EdiThreads=2 will help or harm.

Cu Selur

[maudit]

just by adding the line
PreFetch(3) or PreFetch(4)
at the end of the script, let it works much better, the cpu load is now around 100%. thanks

so, I no longer have to use that "EdiThreads=Y"?
what does the prefetch command do in few words?
it shuould be placed just after the QTGMC line or is the same wherever I put it?
why the parameter "3" is good for a 4x2 core cpu? is it?

[Selur]

what the prefetch command do in few words?

->

Prefetch (clip, int “threads”, int “frames”)


clip =

Input clip.


int threads =

Number of threads to use. If it is 0, it passes without doing anything.
default: (number of logical cores in the system) +1


int frames =

Number of frames to prefetch. Again, if it is 0, it passes without doing anything.
default: threads * 2

source: http://avisynth.nl/index.php/SetFilterMTMode#Prefetch (I usually use prefetch with the number of physical cores)

[maudit]

I have already read it, thx!
anyway also using the command alone with no parameters

Prefetch

it works good, I presume it used the default value:
default: (number of logical cores in the system) +1
that should be 9 in this case.

so what is the best way to use it with a 4 physical cores and 8 threads cpu?
Prefetch 3,4,7,8,9?

[Selur]

There is no 'best' way, it always depends on the filters you use and your system.
-> use AVSMeter to find the setting that best suits your setuo and script.

[VoodooFX]

just by adding the line
PreFetch(3) or PreFetch(4)
at the end of the script, let it works much better, the cpu load is now around 100%. thanks

what does the prefetch command do in few words?
it shuould be placed just after the QTGMC line or is the same wherever I put it?
why the parameter "3" is good for a 4x2 core cpu? is it?

Pefetch enables "MT." You can put it at the very end or you can put it after QTGMC, depends how you want that MT to work. Btw, not all filters are "Prefetch friendly".

For optimal use you should leave some cores "free" for encoder and other programs.

[maudit]

For optimal use you should leave some cores "free" for encoder and other programs.

I presume you are meaning "cores" as "physical core" (4-1), and not "cpu threads" (8-1).
If yes it's ok, I got it. I'll use the Preftech(3) command next to QTGMC or at the spript end, now I see if there are differences or not.

Thank you both Voodoo and Selur, very much for your quick help.

I think I'm fine.

[VoodooFX]

I meant virtual cores included.

[maudit]

I meant virtual cores included.

ok then, I'll use Prefetch(7) or Prefetch(6).
thx

[VoodooFX]

It all depends from many factors, if you get 100% CPU usage with Prefetch(3), then 3 is enough.

[maudit]

It all depends from many factors, if you get 100% CPU usage with Prefetch(3), then 3 is enough.

I see, I just finish trying multiple times all the possible combinations:
Prefetch(3), or 7 or default, in every possible way, at the script end, or next to QTGMC.

is all almost the same, close to 100% cpu load and close to 2.0ghz.

I'll use Prefetch(3) next to QTGMC.

thanks

[jagabo]

I'll use Prefect(3) next to QTGMC.

How many threads gives optimal throughput will also depend on what encoder you are using. When using a CPU based encoder (which looks like what you are doing now) both the encoder and AviSynth will be using CPU time (and threads). But if you switch to using a GPU based encoder prefetch(3) will probably not be enough.

[poisondeathray]

CPU "usage" alone is not necessarily a good metric; you can get CPU thread "thrashing" or wasted cycles. Higher CPU usage does not necessarily mean faster output. You want to look at CPU usage in combination with FPS

Likely you can make that script slightly faster by applying a vertical mod4 crop before QTGMC (QTGMC is usually the bottleneck, and fewer pixels to process is generally faster)

FFVideoSource("\title_t00_track1_[eng].mkv", fpsnum=25, fpsden=1, threads=1)
crop(0, 72, -8, -72, true)
QTGMC(Preset="Slow").SelectEven()
crop(0, 2, 0, -2, true)
FillBorders(0,1,1,0)
bbmod(2,1,2,2,128,8)
prefetch(4)