3 ms·
On my 128GB M4 Max Mac Studio, generating a 15s 480p video with MiniMax H3 in ComfyUI takes an hour and a half. Put Codex to work on deploying it now, hoping t
by linzhangrun 2mo ago
On my 128GB M4 Max Mac Studio, generating a 15s 480p video with MiniMax H3 in ComfyUI takes an hour and a half.
Put Codex to work on deploying it now, hoping the speed can improve quite a lot :-) Thanks anyway
- sscarduzio 2mo agoPlease keep up posted about the results!
- linzhangrun 2mo agoFirst batch of quick test results: approximately 1/5 speed improvement
- bewareofscams 2mo ago+20% or x5 speed improvement?
- linzhangrun 2mo ago20%
- embedding-shape 2mo ago> On my 128GB M4 Max Mac Studio, generating a 15s 480p video with MiniMax H3 in ComfyUI takes an hour and a half. That's crazy, a RTX Pro 6000 does that in in 2-3 minutes (give or take, depending on your exact settings). LLMs don't make the difference between standalone GPU vs unified memory + CPU so obvious as diffusion models seems to do.
- whywhywhywhy 2mo agoIt’s always been the case, it’s more the anomaly that LLMs work at comparable speeds on M series because almost all other ML runs way faster on Nvidia cards.
- MrScruff 2mo agoLLM prompt processing and diffusion models are compute bound, while LLM token generation is memory bandwidth bound.
- throw1234567891 2mo agoAn RTX6000 is a completely different class of hardware.
- embedding-shape 2mo agoReally? No wonder I keep trying to type on it like a laptop but it doesn't work and doesn't even have a display!
- throw1234567891 2mo agoRight? Do you also type on a Mac Studio without a keyboard plugged in? Like tap the ethernet port 3 times in a row then this sequence of sticking your fingers into TB5 ports? I mean, it’s clear you stick your RTX into a computer.
- brcmthrowaway 2mo agoFor gods sakes, Apple let people have run other GPUs instead of these pissweak 2012 class mobile GPUs
- mandeepj 2mo ago> Put Codex to work on deploying it now Which codex?