← Back to community benchmarks

rnj-1-instruct

M3 Ultra (80c) · 512 GB · 8bit · 2026-03-14
Performance
32k
tokens
665.2
PP tok/s
22.4
TG tok/s
49262
TTFT (ms)
17.9
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.11
macOS macOS 26.3
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,027 56.6 10.0 GB view
4k 1,094 50.1 10.9 GB view
8k 1,013 41.9 12.0 GB view
16k 869.6 31.3 13.9 GB view
32k 665.2 22.4 17.9 GB current
Batching Results
Batch Size TG tok/s Speedup
56.6 1.00×
99.1 1.75×
148.2 2.62×
184.5 3.26×