I noticed that in Task Manager my GPU usage when upscaling is only between 10-20%. In afterburner it is 20 to 40%. CPU usage is around 50% which I guess is because the app is limited to 2 processes in the perferences (not delibaretely). I’m getting around 4fps with that for processing a 720p30 video to 1080p60 via chronos fast and Artemis Low Quality, Noisy, encoded to ProRes 442 HQ.
Is there something I could do to utilize the GPU more?
A few lines I picked from the logs in case it helps:
WINDOWS VERSION IS 11.2009
VNNI: false AVX: true CPUName: AMD Ryzen 7 5800X 8-Core Processor Thread Count: 16
MEM TOTAL: 31.9235 AVAIL: 23.0991
Device Id: 0 Name: NVIDIA GeForce RTX 3080 Ti Cores: 0 VRAM: 11.8242 GB Used VRAM: 0 GB Free VRAM: 5.66211 GB DataType: 2 Serial: 012a75 ComputeLevel: 806 Visible: true Legacy: false
And my Nvidia info:
Operating System: Windows 10 Pro, 64-bit
DirectX version: 12.0
GPU processor: NVIDIA GeForce RTX 3080 Ti
Driver version: 516.94
Driver Type: DCH
Direct3D feature level: 12_1
CUDA Cores: 10240
Resizable BAR No
Core clock: 1665 MHz
Memory data rate: 19.00 Gbps
Memory interface: 384-bit
Memory bandwidth: 912.10 GB/s
Total available graphics memory: 28632 MB
Dedicated video memory: 12288 MB GDDR6X
System video memory: 0 MB
Shared system memory: 16344 MB
Video BIOS version: 94.02.71.40.49
IRQ: Not used
Bus: PCI Express x16 Gen4
I have an RTX 3080ti and a 5900X. I’ve never done 720 to 1080 in TVAI 3.0.1, but in VEAI 2.6.4 that was always about twice as slow as 480 to 1080, and resulted in worse looking videos. Doesn’t make sense to me.
I’d say your GPU and CPU usage look normal to me. It’s kind of odd to me that if I have one instance running, GPU usage is at 30%, but if I run two it jumps up to 90%.
Enabling Resizable BAR might get you a little boost. I’ve heard people say system RAM can effect performance, but I’ve never seen anyone back that up with evidence.
Looks reasonable to me too. Adding frame interpolation slows things down quite a bit since it needs to process a rolling block of frames. You might try the latest Nvidia driver (you are a few releases behind) - Topaz requires at least 522.25 (game ready driver) or 522.30 (Studio driver) starting in 3.0.1
Isn’t that only if you try to use the NVENC output?
Oh, I guess I had already updated to 522.30 a few weeks ago. For some reason I thought it was a much newer driver.
CPU at work is about 50%, GPU 30%, RAM and VRAM usage is low and plenty left over.
Why is single-tasking so slow when resources have room to do additional tasks?
My PC is an i7-4790 RAM 32GB RTX2060 (12GB) SSD, so it’s not the latest, but I can’t understand why the speed is slow even though there are extra resources.
Reading the report, it seems that even those who use the latest PCs have excess capacity.
Why is it like this?
Is this something that can’t be improved with software improvements?
As far as I can tell it’s the tensor cores actually. Because they are what is being used, none of the monitoring tools know how to show their usage. My GTX 1060 gets 100% used on the CUDA graph in Windows Task Manager. With 2.6.4 it was graphics_1. Both those graphs don’t exist on my RTX 3080ti. I’m guessing they’re all merged into the 3D graph. It could be using 100% of the Tensor cores for all we know.
What ForSerious said: the load readings might be inaccurate. I think I had other apps showing little load too but the card was actually working hard with power consumption at 350w. You may want to use HWInfo to check how much power it actually draws because that reading seems to be correct.
