Both are on your vendor's spec sheet. On a Mac, use about three quarters of your memory.
Popular
Pick your machine on the left, or start with one of these.
Pick a model on the left to rank every machine by speed.
Popular comparisons
Build your own- DGX Spark (GB10) vs M4 Max · 128 GB · 40c103 vs 101 models run well
- DGX Spark (GB10) vs RTX 5090103 vs 86 models run well
- DGX Spark (GB10) vs Ryzen AI Max+ 395 · 128 GB103 vs 101 models run well
- DGX Spark (GB10) vs M3 Ultra · 512 GB · 80c103 vs 109 models run well
- DGX Spark (GB10) vs RTX PRO 6000 Blackwell103 vs 101 models run well
- RTX 5090 vs RTX 409086 vs 80 models run well
- RTX 4090 vs RTX 309080 vs 80 models run well
- M4 Max · 128 GB · 40c vs M3 Ultra · 256 GB · 80c101 vs 105 models run well
A model fits when its weights, plus a margin for the runtime and an 8K-token context, fit in the memory your machine gives it. Speed is capped by memory bandwidth, because each new token reads the active weights once. Method and sources