Performance
8k
tokens
663.9
PP tok/s
25.9
TG tok/s
12339
TTFT (ms)
229.1
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.2.6
macOS
macOS 15.5
Context
8,192
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 600.4 | 34.2 | 227.2 GB | view |
| 4k | 707.9 | 30.1 | 228.2 GB | view |
| 8k | 663.9 | 25.9 | 229.1 GB | current |
| 16k | 564.2 | 21.1 | 231.1 GB | view |
| 32k | 427.8 | 14.6 | 235.0 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 34.2 | 1.00× |
| 2× | 48.7 | 1.42× |
| 4× | 63.2 | 1.85× |
| 8× | 99.8 | 2.92× |