antirez hits 25 t/s running DeepSeek v4.1 Flash across two MacBooks via RDMA
antirez · x · 2026-09-11
antirez continues experimenting with DeepSeek v4.1 Flash on MacBooks:
- Tensor parallelism at 50% per machine over RDMA yields 25 tokens/s, vs 15 t/s with SSD streaming on a single machine.
- He notes the model's decoder is unused during prefill, so on a 128GB machine all layers for large prefills can be fully resident in memory.
- He sees significant optimization headroom for this use case.
More from Infra
- Anthropic S-1: only $19B total compute spend since inception, overbuild warning — bookwormengr · 2026-10-27
- CoreWeave, You.com and NVIDIA lift BrowseComp 22.9% in 3-week RL post-training sprint — RichardSocher · 2026-10-27
- Grid queues make power architecture an AI-factory decision; Exowatt and Redwood offer fixes — ingliguori · 2026-10-27
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27
- AI Infra Meme: While Countries Need TW-Scale Power for Datacenters, People Argue About Lightbulbs — mgostIH · 2026-10-27
- NVIDIA reportedly stops taking RTX 5090 orders from board partners, even backorders — chemist_slime · 2026-10-27