I needed more VRAM. Things escalated.

Sep 21, 2026 · 5:24 PM UTC

6
22
1,774
It works. Llama 3.3 70B loaded across both GPUs.
1
3
244
Sort replies: Relevant Recent Liked
Replying to @smallfly
No hate! No shame! 😳
1
1
85
Whatever it takes 😅
1
58
Replying to @smallfly
been trying to buy second hand 3090s lately, they're rare and expensive, so - I'm a bit jealous but - enjoy your compute !! having a dedicated GPU to run a model while still being able to work is very comfortable :)
1
2
141
I know! I kept that one in its box for a while knowing I’d use it eventually. And you’re absolutely right. More VRAM isn’t just about bigger models, it’s also about having several models loaded at once, or still having some VRAM available while one is running. Avoiding the constant load/unload pain is a big part of it.
2
46
Replying to @smallfly
I wonder how the RGB would splat!
1
1
58
Nicely, am I quite sure.
56
Replying to @smallfly
Holy guacamole 🥑 @smallfly
1
1
56
It works. Now I need to clean it up 😅
1
41
Replying to @smallfly
1
6