M4 Max Benchmarks: oMLX Wins at Long Context, Prefix Caching is Key
vitordeas · reddit · 2026-08-30
Benchmarked 5 runtime/quant stacks running Qwen3.8-27B (32K-256K) on an M4 Max (128GB).
Key Takeaways:
- ≤128K: MTPLX 4-bit (native MTP) is the throughput king.
- 256K: MTPLX verification collapses (7 tok/s); oMLX and mlx-dspark lead (14-15 tok/s).
- Biggest Speedup: Warm prefix cache cuts 256K prefill from 40 mins to 2 mins. oMLX supports edit-in-the-middle caching; MTPLX does not.
Tech Specs: Compares oMLX (AWQ/oQ8e), MTPLX (native MTP), and mlx-dspark (DFlash2 drafter).
More from Infra
- Copper to optical: optical backplanes seen as necessary as racks move to optics — BenBajarin · 2026-10-16
- Polymarket prices 28% odds a US state enacts data center moratorium by year-end — Polymarket · 2026-10-16
- Pennsylvania residents reject $10,000-per-household offer for massive Amazon data center — Polymarket · 2026-10-16
- PyTorch now runs natively on AWS Trainium with zero code changes — PyTorch · 2026-10-16
- Chinese labs run on ~1/10 US compute; ByteDance pays 2-4x to hoover up SEA supply — robleclerc · 2026-10-16
- Zymtrace raises $12.2M and hires ML engineers to build agents that fix GPU bottlenecks — cfregly · 2026-10-16