← Back to community benchmarks

gemma-3-4b-it-qat

M1 Ultra (48c) · 64 GB · 8bit · 2026-03-15
Performance
32k
tokens
1,457
PP tok/s
43.3
TG tok/s
22487
TTFT (ms)
7.3
Peak mem (GB)
Hardware
Chip M1 Ultra (48c)
Memory 64 GB
GPU Cores 48
Software
oMLX v0.2.13
macOS macOS 26.3.1
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,270 28.9 6.0 GB view
4k 1,493 81.6 6.3 GB view
8k 1,551 74.9 6.5 GB view
16k 1,529 63.3 6.7 GB view
32k 1,457 43.3 7.3 GB current
64k 1,273 27.3 8.5 GB view
Batching Results
Batch Size TG tok/s Speedup
28.9 1.00×
32.7 1.13×
48.9 1.69×
70.4 2.44×