How Much VRAM ?

#7
by Ni3SinghR - opened

How much VRAM Required to run this model?

  1. You can run it with 750gb of RAM(not VRAM).

I don't think that's necessary.. You can run it on 400GB RAM when quantized. You'll need about 96GB VRAM to be able to run it decently fast.

  1. You can run it with 750gb of RAM(not VRAM).

9eb1f745b8c5cd23db4a2316daa39e71

Rule of thumb , for fp8 you will need around the same amount of parameters for vram,
for example :
70b -> ~ 70 gb vram
30b ->~ 30gn vram

For fb16 it will be twice
70b ->~ 140gb
30b->~ 60gb

Sign up or log in to comment