← Back to community benchmarks
gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking-mxfp8
Performance
32k
tokens
1,950
PP tok/s
48.0
TG tok/s
16801
TTFT (ms)
10.1
Peak mem (GB)
Hardware
Chip
M4 Max (40c)
Memory
128 GB
GPU Cores
40
Software
oMLX
v0.3.12
macOS
macOS 26.5
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 2,153 | 62.4 | 8.9 GB | view |
| 4k | 2,131 | 61.1 | 8.9 GB | view |
| 8k | 2,247 | 58.4 | 9.1 GB | view |
| 16k | 2,147 | 54.5 | 9.4 GB | view |
| 32k | 1,950 | 48.0 | 10.1 GB | current |
| 64k | 1,648 | 37.3 | 10.9 GB | view |
| 128k | 1,337 | 26.8 | 13.2 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 62.4 | 1.00× |
| 2× | 101.9 | 1.63× |
| 4× | 142.3 | 2.28× |
| 8× | 154.9 | 2.48× |