What is the VRAM requirement to run this ?

#1
by RageshAntony - opened

How much VRAM I need to run this ? Is A100 (80GB) is enough ?

Listed here, depends on config:

https://unsloth.ai/blog/deepseek-r1

@lilmildred

I read it already. Under the "Hardware Requirements for R1 / Zero:" , only CPU based run with requires 48GB is mentioned. What about GPU based ?

@lilmildred

I read it already. Under the "Hardware Requirements for R1 / Zero:" , only CPU based run with requires 48GB is mentioned. What about GPU based ?

Do you really think "bit" as the unit of information different on CPU or GPU? It's same and the volume to fit them are same on CPU & GPU (you only choosing where to offload it - in system RAM or in VRAM, or in both and size cannot be reduced), so i think Q5 will need a 512Gb GPU card or sum of GPU cards VRAM, because on CPU Q5 takes 502Gb to load in V3 which is same size model (i can tell later precisely when will test it). There's "enterprise" (corporate) level motherboards with a terabyte RAM in DDR4 or DDR5. Speed not important, the volume to start it is the first goal and the barrier today (consumers will not receive big volumes by presentations on CES, max 256Gb, that's industry decision).

@krustik
I am new to AI world. That's why I asked like this.

One more question. Is it possible to run this R1 in AMD MI300X ?

image.png

It's just $12 (and $7 for spot) at RunPOD ?

Unsloth AI org

@krustik
I am new to AI world. That's why I asked like this.

One more question. Is it possible to run this R1 in AMD MI300X ?

image.png

It's just $12 (and $7 for spot) at RunPOD ?

um yes definitely

Sign up or log in to comment