Hi all,
I’m testing the Iris-2 model using Topaz Video AI CLI version 6.1.2 on a Mac Mini M4 with 16GB RAM. I’ve noticed some unexpected memory behavior during upscaling:
-
At a resolution of 7168×3584, RAM usage is around 3GB and the process runs smoothly.
-
At 8192×4096, (8K) RAM usage skyrockets to 20+GB, causing the system to run out of memory.
-
This occurs even when I set -filter_threads to 1 or 6. I also tried vram=0.1 and instances=1, but the effect is none. I tried more different settings.
-
I’m using the CLI exclusively, and I’ve seen the same behavior in higher versions as well.
Examples of “tvai_up”:
RAM Problem:
“tvai_up=model=iris-2:scale=0:w=8192:h=4096:preblur=0:noise=0:details=0:halo=0:blur=0:compression=0:estimate=6:blend=0.5:device=-2:vram=0.10:instances=1,scale=w=8192:h=4096:flags=lanczos:threads=0”
No problem:
“tvai_up=model=iris-2:scale=0:w=7168:h=3584:preblur=0:noise=0:details=0:halo=0:blur=0:compression=0:estimate=6:blend=0.5:device=-2:vram=0.10:instances=1,scale=w=7168:h=3584:flags=lanczos:threads=0”
Also ffmpeg option “filter_threads” seems to be ignored.
It seems like the memory usage is highly non-linear with resolution, possibly related to internal tile sizes of the Iris-2 model? I haven’t been able to find official documentation on tile sizes or how the model manages RAM.
Has anyone else observed similar behavior, or can provide insight into Iris-2 tile sizes or internal memory management?
Thanks in advance!
