DeepSeek 4.1 Flash Hits 200 TPS Full Precision on 4 Max-Qs with Just 64GB RAM
TheZachMueller · x · 2026-09-11
A developer ran DeepSeek 4.1 Flash at full precision with the DSpark engine on 4 Max-Q chips with only 64GB of system RAM, reaching 200 TPS.
The key trick: offloading the 200GB Engram (essentially a hash table) to NVMe instead of keeping it in RAM works surprisingly well and saves cost. The author says there's more headroom and a vLLM recipe is coming.
More from Infra
- DeepSpeed shows AutoEP/AutoTP/AutoSP parallelism gains, hosts office hours — PyTorch · 2026-10-27
- Cloudflare Durable Objects validated as top architecture for AI agent WebSocket servers — tristanbob · 2026-10-27
- Gary Marcus: With Q3 data added, Anthropic's compute-efficiency growth shows deceleration — GaryMarcus · 2026-10-27
- SpaceX locks in $40B GPU financing deal with Apollo, the largest ever committed — TechCrunch AI · 2026-10-27
- Inference is just Kubernetes with a multi-trillion-dollar market, and the communities overlap — matt_slotnick · 2026-10-27
- vLLM eyes Rust migration as inference engines race to rewrite the stack — AccBalanced · 2026-10-27