Comment by garciasn

We run our LLM workloads on a M2 Ultra because of this. 2x the VRAM; one-time cost at $5350 was the same as, at the time, 1 month of 80GB VRAM GPU in GCP. Works well for us.