← Back to community benchmarks
Jundot-Qwen3.6-27B-oQ8-mtp
Performance
195k
tokens
158.1
PP tok/s
15.3
TG tok/s
1264775
TTFT (ms)
68.9
Peak mem (GB)
Hardware
Chip
M3 Ultra (80c)
Memory
512 GB
GPU Cores
80
Software
oMLX
v0.3.9.dev1
macOS
macOS 26.3.1
Context
200,000
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 384.2 | 27.8 | 28.8 GB | view |
| 4k | 417.3 | 30.4 | 30.3 GB | view |
| 8k | 417.7 | 30.3 | 31.3 GB | view |
| 16k | 407.7 | 28.7 | 32.8 GB | view |
| 32k | 371.3 | 26.4 | 35.8 GB | view |
| 64k | 232.9 | 17.6 | 41.8 GB | view |
| 128k | 185.5 | 18.2 | 54.2 GB | view |
| 195k | 158.1 | 15.3 | 68.9 GB | current |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 27.8 | 1.00× |
| 2× | 31.8 | 1.14× |
| 4× | 46.6 | 1.68× |
| 8× | 54.2 | 1.95× |