I am using a VapourSynth setup (Windoze), with a pipe to x265. I had found a ffmpeg way of processing png files, like -i %06d.png to the parameters, but I can’t have my cookie and eat it, as I also need the -i for something like -f vapoursynth -i “f:\jobs\test.vpy”. Will take a bit of sorting out.
See, what I don’t get, is that my CPU (i9 12900k) is already at near full load, keeping my graphics card (RTX 3080 Ti) busy for like only 20%. My point being, what would a faster card do, when the CPU is already at capacity? (Aka, cannot process the GPU data faster)
What software do you use to monitor the load of the gpu.
Here is the comparison between MSI Afterburner and Taskmanager for the workload detection.
The task manager does not even show half of the half although the GPU is fully loaded.
i did post this whyle the 3.0.5 beta.
The taskmanager did readout every time the gpu was at idle load between the frames.
But you need to set the afterburner to 100ms readout.
I normally use the Task Manager; GPU-Z gives yields slightly higher values, with peaks to 86%, but not sustained.
This is normal, the video is always copied to the gpu in single frames and the GPU needs less than half a second to process the image.
GPUZ also does not read out fast enough to show the peaks.
Thats why you don’t see any grow in performance with a new gpu.
Makes sense. Next time, I will try afterburner.
N.B. They changed the forum layout right from under me, while I was typing. Like the upgrade to TVAI, it’s no improvement (like I can no longer click on to see who liked something).
I’ll be very curious to know if the team is working on ways the software can accelerate using the new features in the RTX 4000 series. Got the 4080 and it’s not impressing at this point even though I know it’s powerful in theory.
For example, what is Video AI take on using CUDA? What about Tensor Cores (now using a new more efficient architecture)? Or the new DLSS 3?
Each of these would have practical implications that can speed up processing but is the team actively looking at developing features that can be accelerated with this hardware? Are we only seeing improvement with memory speed / Ghz or are we taking advantage (or working towards that) of those specific technologies?
I’m pretty sure the ‘prob-3’ is just referring to Proteus model. (not sure on the letter ‘b’)
For instance, Gaia - High Quality (HQ) model will print out: ‘ghq5’ or similar.
We’re here at image and Video Processing.
So no DLSS is used since this is not Gaming.
Tensorcores are used but it doesnt matter if your frame is finished in 0,0002 or 0,0001 sec but 0,0001 needs 2000 teraflop and 0,0002 needs 1000 teraflop since its half as fast.
You get my point?
but here are some other examples
Its the same with TVAI at the moment.
So we have to wait for devs to improove the software.
I think I get your point, basically - it’s down to teraflops, thanks for the reply!
Still, would be great to hear about what they are planning going forward in terms of optimization of the software for the available hardware.
CUDA is nVidia-specific, they’ll never use that (also not sure it would really help, tbh).
DLSS 3? Wut? That’s an A.i upscaling technique (also nVidia-specific); and yes, it’s also using A.i, but no good for TVAI in any way.
We betatesters don’t know that either at the moment.
As you can see with video editing, it probably takes a fundamentally different approach to see more than a 50% performance increase.
At the moment the CPU is more important.
I think fundamentally, the Topaz team need to communicate better as to what is going to drive performance up and what is likely to create a bottleneck (inc. if they are working on any features that can be sped up using different hardware). What you say makes perfect sense, which is a pity as I was very convinced investing in a new faster GPU is going to make a massive difference.
They make great software, nobody will deny that, but they need to cater a bit better to explain how the experience is going to be with different hardware and what the end user can do on their end to make using the software a nicer (faster) experience.
At the start I want to point out that I used versions 2.4.0 and 2.6.4 for testing. I haven’t purchased a 3.0 license yet, I believe it is still in beta. After moving from 3090 to 4090 there was no difference in performance. The GPU was bored and was drawing 70-90W. Back then I had a Ryzen 5800X processor. Now I changed platform to AM5, 7950X (105W ECO mode). And there was a speed increase of 33-40% in the Proteus model (1080p to 4K) and almost 100% in the chronos model (24fps to 60fps in 4K). Still not much considering the cost. ;/
I’m having a problem. Whilst attempting to stabilize a long video, when it get near the end of the job the log shows an input file read error and the whole thing deletes and TVAI quits, leaving nothing but wasted time. I have been storing my work on a Seagate 8TB drive with several TB free. - I uploaded the logs to the Dropbox “PDWhite Logs runs into errors at end Stabilization Crashes ForSupport.tar”
This has recently happened several times attempting various Export operations, both in v3.0.6 and in the latest beta, as well.
I am going to move my input file to an empty 2TB SSD and see if that makes any difference.
40% is significant.
Thank you for the post.
Replying to my own post. I moved the same files that I have been crashing on, logged as read errors when they were being run from the Seagate 8TB Hard drive. I am presently running the Stabilization routine source and destination on a nearly empty 2TB NVMe SSD drive. So far there appears to be no problem and the job has already passed the point where it aborted due to the file error working off the hard drive. - This makes me suspect that the problem may be stemming from a read or write timeout situation.
Has anyone else noticed a similar situation?
Is that an external or internal drive? I have a 16tb external Seagate and the startup after timeout is terrible at almost 10 seconds. Plex sometimes throws a "file not found’ error. I had a little script years ago when I used a lot of externals for normal use that would go out and write and over write a small text file every 10 minutes to keep the drives spun up.
Go into your power management settings and set the hard drive sleep time to a large number. Screenshot for reference only.
New release available: v3.0.7