← Back to community benchmarks
gemma-4-31B-it-PARO
Performance
32k
tokens
236.5
PP tok/s
21.0
TG tok/s
138535
TTFT (ms)
24.9
Peak mem (GB)
Hardware
Chip
M3 Ultra (60c)
Memory
256 GB
GPU Cores
60
Software
oMLX
v0.3.9
macOS
macOS 26.5
Context
32,768
Performance by Context Length
| Context | PP tok/s | TG tok/s | Peak Mem | |
|---|---|---|---|---|
| 1k | 267.0 | 30.4 | 17.0 GB | view |
| 4k | 267.6 | 29.4 | 19.7 GB | view |
| 8k | 267.3 | 28.0 | 20.5 GB | view |
| 16k | 256.4 | 25.7 | 21.7 GB | view |
| 32k | 236.5 | 21.0 | 24.9 GB | current |
| 64k | 204.0 | 15.1 | 30.9 GB | view |
| 128k | 157.3 | 11.1 | 42.9 GB | view |
Batching Results
| Batch Size | TG tok/s | Speedup |
|---|---|---|
| 1× | 30.4 | 1.00× |
| 2× | 44.5 | 1.46× |
| 4× | 52.4 | 1.72× |