Apple M2 Max 64GB Llama-3.3-70B 4bit token-gen
8.8 tok/s 4bit (LM Studio)
Fields
id
benchmarks/comm-apple-m2-max-llama33-70b-4bit-tgMetric
decode_tok_s
Value
8.8
Unit
tok/s 4bit (LM Studio)
Model
Llama-3.3-70B
Methodology
Silicon Score Llama-3.3-70B model page row; M2 Max 38-core, 64 GB, 4-bit via LM Studio; ~400 GB/s unified.
Measured by
third_party
Thermal state
unknown
notes
64 GB config only fits a 70B Q4 at LOW context - capacity, not bandwidth, is the constraint here. Bandwidth-scaled from the same table: M1 Ultra (800 GB/s) 12.6 t/s, M3 Ultra (800) 15.5, M4 Max (546) 11.8 - M2 Max at 400 GB/s landing 8.8 is consistent with the bandwidth-limited decode model.
Notes
64 GB config only fits a 70B Q4 at LOW context - capacity, not bandwidth, is the constraint here. Bandwidth-scaled from the same table: M1 Ultra (800 GB/s) 12.6 t/s, M3 Ultra (800) 15.5, M4 Max (546) 11.8 - M2 Max at 400 GB/s landing 8.8 is consistent with the bandwidth-limited decode model.