DeepSeek slashes KV cache to 890 bytes/token, hinting at million-agent swarms
teortaxesTex · x · 2026-09-10
Analysis of DeepSeek V4.1 Flash's engineering: KV cache compressed to 890 bytes/token (<1GB per 1M tokens), cutting HBM requirements 3.9x and SSD needs 8x versus V4-Flash — half a year on, no other model approaches V4 arch's KV efficiency. On a 96GB/950DT minimal setup, the author estimates 1.3 million max-budget agents could run in parallel on the Huawei cluster acquired in May. Margins reportedly exceed 90%; 'if this were an American lab, we'd be talking about RSI.'
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
- 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
- Resin lets an agent build standalone LLM inference engines; vLLM and SGLang optimize DeepSeek kernels differently — AccBalanced · 2026-10-27