Apple M3 Ultra (512GB) Llama-3.3-70B 4-bit decode
15.5 tok/s (4-bit, 8K ctx)
Fields
id
benchmarks/comm-apple-m3-ultra-llama33-70b-4bit-decodeMetric
decode_tok_s
Value
15.5
Unit
tok/s (4-bit, 8K ctx)
Model
Llama-3.3-70B 4-bit
Methodology
Silicon Score measured table; 8K context; ~800 GB/s unified memory
Measured by
third_party
Thermal state
unknown