← Back to community benchmarks
FP8_converted
Performance
4k
tokens
722.0
PP tok/s
32.2
TG tok/s
5673
TTFT (ms)
222.7
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.12b1
macOS
macOS 26.3
Context
4,096
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 596.2 | 36.1 | 221.6 GB | view |
| 4k | 722.0 | 32.2 | 222.7 GB | current |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 36.1 | 1.00× |
| 2× | 50.2 | 1.39× |
| 4× | 70.3 | 1.95× |