I am trying to decide best way to run Topaz Video with Starlight locally. I need to upgrade, but I don’t know which renders best. Is Starlight focused on the amount of ram I have like a typical LLM? Right now I am getting about 1 minute per hour with my 3090.
I am thinking of a dedicated Mac Studio with 128gb or ram, or a server with a 4090 in it to run starlight full time….
Smack some knowledge on me please.
4090 (or maybe even something like 5070Ti upwards).
AppleSilicon is much slower (more like 4060-4070 performance) and to make things worse Starlight doesn’t even work correctly on that platform.
No one knows when (if?) Topaz get their act together with ARM support and at what performance…
I have tried running it on a Mac Studio M2 Ultra with 64mb. I tried a 2X upscale of a 1440 X 960 video It ran at 0.1 FPS and produces unusable garbled output. Tried it at no upscale and it is currently running at 0.3 FPS and is estimating a completion time of over 4 days for a 45 minute video.
See comment like this confuse me because though the 5070ti is a generation newer, it has 8gb less ram; 16 vs 24. So is it ram or tops or what?
You can cancel that run and spare your time. The output will be garbled as well.
Starlight just does not work on AppleSilicon atm with macOS 26.2.
Well, it’s a bit of both - but in my experience 16GB of RAM are enough and for the speed you desperately need those TOPS.
(Even the 12GB VRAM in my 4070 are enough here to do SD → FHD or even 3x upscales)
