When switching my device to “CPU” in the current version of Video Enhance AI, I notice that the program only makes use of 64 threads / logical processors. This is an issue, because it means CPU operations take twice as long as they should on my system, since the program is only using half of the available resources. My CPU is an AMD Epyc 7742 (64 cores / 128 threads with SMT enabled) and I am using Windows Server 2022 Datacenter.
The 64 thread issue I am experiencing is a common limitation in Windows which manifests in legacy programs which are not processor group aware (more information on this is available from Microsoft’s website here).
I am wondering if there are any plans to remedy this limitation in an upcoming release of VEAI?
viktorz3008
(Z...z...z 😴 I'm sleeping. And I love Sumire 😍.)
2
The program uses GPU the most. CPU with high core count will NOT help. Single clock speed matters here. It seems that you use CPU mode which is much slower than GPU mode. So that’s the reason. This is not Blender, so anything more than 8 cores will not benefit. You should get a desktop CPU and a powerful GPU.
The program uses 100% of exactly 64 logical processors. This makes it clear it is a processor group limitation, not a lack of load.
viktorz3008
(Z...z...z 😴 I'm sleeping. And I love Sumire 😍.)
4
You don’t understand what I was saying. If all the cores are used, it means you’re using CPU mode without a GPU. And CPU mode is super slow compared to GPU. Even 128 cores or what, GPU is more suitable for this job than CPU.
Anyway VE AI does not use the full power of the GPU. Sometimes running two or even three instances of VE AI to work simultanously via GPU speeds up the process a lot!
viktorz3008
(Z...z...z 😴 I'm sleeping. And I love Sumire 😍.)
8
Im my testing, 480 => 1080 (maybe)
But 1080 => 4K, not really.
Is this still a thing? Considering interesting AMD CPUs with 96 or even 128 cores, respectively 192 and 256 threads. What actually caps this at 64 threads in the app (the programming framework)?
It’s moved to being an FFMPEG filter. You can lookup to see if FFMPEG has that limit. I think it usually uses what the operating system says is available.
That being said, TVAI favors a powerful CPU GPU combination. I’ve been telling people that it needs at least 10 powerful CPU cores, but it might be as low as 8.
Unless you’re planning on doing all the processing on the CPU, you’d probably need like 3 to 4 RTX 4090s to use all of the cores on that CPU. And you’d need to be running about 3 instances of TVAI per GPU—so 9 to 12 total. Storage would probably become a bottleneck at that point. Oh and when I say instances of TVAI, that means they’re each processing a different movie. I think we’re still years away from TVAI being redesigned to run on a render farm.