Strata revives decade-old Maxwell GPUs: local LLM at 150 tk/s prefill on M6000/M40
jmuff98 · reddit · 2026-10-17
A Reddit user ran a local LLM on a Quadro M6000 + Tesla M40 (two Maxwell GPUs, 48GB combined VRAM) using Strata, measuring 150 tk/s prefill and 25 tk/s decode. Not fast or power-efficient, but surprisingly usable — all MoE experts fit in VRAM, with Strata used purely to maximize prefill speed. No code optimization was done beyond porting NVIDIA architecture support so Maxwell cards can run Strata.
More from Infra
- Kimi K3 three months later: 1.56TB is too big to run locally and rivals have caught up — Sanbi_Ai · 2026-10-17
- Epoch AI chart: xAI owns fewer GPUs than CoreWeave, far behind hyperscalers — bookwormengr · 2026-10-17
- Nvidia CEO Jensen Huang says AI stock selloff is temporary: "All of this is going to pass" — Polymarket · 2026-10-17
- Cloudflare rolls out full ZDR on Workers AI: no data retention, no training on your data — michellechen · 2026-10-17
- Modal explains its speed moat: ImageFS tiered filesystem now tricking VM ext4 — dejavucoder · 2026-10-17
- AWS built a production eval framework on Bedrock AgentCore: agents fail silently — AWS ML Blog · 2026-10-17