All models extremely slow / ffmpeg high cpu

i am using a rtx 5090 which gets utilized to about 10% while a ffmpeg.exe process is hugging 50-70 % of a i9-12900k cpu for proteus and iris models when upscaling a 720p to 4k video.

i remeber this being ~20fps on previous versions, even with my 4090 back then.

what’s going on!?

It likely has to do with the export codec you selected. Some codecs use only the CPU.

Thanks for the tip. H264 with cuda?

just testet intel encoding, does not make a difference

Still horribly slow

H.265 should consume GPU resource more than CPU.

I am stuck - any idea on how to troubleshoot this?

Encoding: Topaz only uses NVIDIA NVENC because it’s free (x-lib costs money), unless they have changed that. So h264/h265 encoding can only run on the GPU, so the cpu setting has no effect (selection box should be removed)

I think this 10% gpu usage comes from the h264 encoding only, because the 5090 is so fast there is no higher load and your Iris/Proteus upscale runs on the CPU only, this explains why you’re cpu usage is that high - why Iris/Proteus does not use (anymore?) the gpu, I don’t know.

i had no issues with gpu usage on (much) older versions. proteus/iris was next to realtime. i hope someone from the @TopazAdam @kyle.topazlabs Topazlabs team could take a look :slight_smile:

Test by exporting using H.265, and check Task Manager while the process is running to see what is consuming so much CPU. There might be another reason for it.

same issue with h265 - it’s the ffmpeg.exe process. I’ve tried with different input codecs (h264, h265, ffv1) - does not make a difference.

also no change with different output codecs and devices. i can offload encoding to integrated intel GPU but still high cpu usage for ffmpeg.exe and only minimal load for the 5090

I think your memory is misleading you. I’ve never experienced Topaz reaching anywhere near 20 fps when going from 720p to 4K.

I’ve been using Topaz since version 4.x, and up until version 7.0 I had a system with a Ryzen 7800X3D, an RTX 4090, and 64 GB of RAM—and I never achieved that kind of speed.

Depending on the settings and video quality, I remember occasionally reaching around 11–12 fps, but those were more like exceptions. Most of the time, it was usually around 7–8 fps.

Even with my new setup—a Ryzen 9950X3D and an RTX 5090—I can barely exceed that, since older models like Iris or Proteus still make very limited use of high-end hardware.

Hm i do see 19fps during these 5-second preview renders. yet the full export is slow as hell.

I am getting 2.5-4.8 fps at the moment and the GPU is not utilized.

Yes, in a short preview. Thats mean nothing. The 30 seconds preview is the only one, that can you give a good example how fast the rendering is.

And yes, in the most cases my 5090 are not ulitilzed to. When you want this, try Starlight Precise 2.6 :slight_smile: