← Back to community benchmarks
gemma-4-31B-it-The-DECKARD-HERETIC-UNCENSORED-Thinking-qx86-hi
Performance
1k
tokens
91.4
PP tok/s
10.6
TG tok/s
11207
TTFT (ms)
28.4
Peak mem (GB)
Hardware
Chip
M2 Max (38c)
Memory
64 GB
GPU Cores
38
Software
oMLX
v0.3.8
macOS
macOS 15.7.4
Context
1,024
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 91.4 | 10.6 | 28.4 GB | current |
| 4k | 91.5 | 10.3 | 31.1 GB | view |
| 8k | 90.8 | 10.0 | 31.6 GB | view |
| 16k | 88.9 | 9.4 | 32.8 GB | view |
| 32k | 84.9 | 8.6 | 35.7 GB | view |
| 64k | 77.8 | 7.2 | 41.7 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 10.6 | 1.00× |
| 2× | 16.0 | 1.51× |
| 4× | 18.5 | 1.75× |
| 8× | 18.7 | 1.76× |