Performance
16k
tokens
119.8
PP tok/s
12.4
TG tok/s
136745
TTFT (ms)
402.7
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.6
macOS
macOS 26.3
Context
16,384
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 185.8 | 17.1 | 391.8 GB | view |
| 4k | 181.9 | 14.0 | 394.1 GB | view |
| 8k | 156.4 | 13.4 | 396.7 GB | view |
| 16k | 119.8 | 12.4 | 402.7 GB | current |
| 32k | 76.4 | 10.9 | 415.4 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 17.1 | 1.00× |
| 2× | 24.2 | 1.42× |
| 4× | 31.0 | 1.81× |
| 8× | 41.2 | 2.41× |