Compatibility check
Can I run Mixtral 8x22B Instruct on Apple M3 Ultra (80-core GPU)?
Yes — it runs.best quant:
Decode speed
21 tok/sest
Usable context
66K
Memory
512 GB
Fits in memory21 tok/sest · 66K ctx
Fits in memory18 tok/sest · 66K ctx
Fits in memory12 tok/sest · 66K ctx