DeepSeek-V4.1-Flash hits 59.5 tok/s on a single M5 Ultra at ~160W
AIFlow_ML · x · 2026-11-21
A hands-on measurement by Justin Kalland, shared by @OwnYourCompute, shows DeepSeek-V4.1-Flash running at 59.5 tok/s on a single M5 Ultra 512GB machine drawing only about 160W.
The takeaway: at home, power is the small part of the cost — the box itself is the real expense. A notable datapoint for running frontier-scale models locally on high-end consumer hardware.
More from Infra
- PostHog cuts LLM costs with decision models: up to 444.6x cheaper, 193.6x faster — yawnxyz · 2026-11-21
- Sarvam resells DeepSeek in India, drawing backlash over AI sovereignty stance — teortaxesTex · 2026-11-21
- Ilya Sutskever warns weak neocloud security could let rogue agents hijack compute — npinto · 2026-11-21
- Colfax and MIT speed up lattice QCD staple networks 700–1200x on H200 with CuTe DSL — snowclipsed · 2026-11-21
- Will MoE Inkling 2 Mini get strata support and rival dense Qwen on 12GB VRAM? — ECrispy · 2026-11-21
- 'Computers Have Broken Everything': NYT on America's growing tech fragility — nordicinst · 2026-11-21