DeepSeek's DSec paper: 3M sandboxes/day, 380K concurrent, powering V3.2-V4.1 RL
jiqizhixin · x · 2026-10-02
DeepSeek (with Liang Wenfeng among 100+ authors) published DSec (DeepSeek Elastic Compute), a 31-page systems paper on its production RL sandbox platform: 160 CPU nodes, 30,000 cores, 250 TB DRAM, PB-scale images and environment layers, 3 million sandbox instances per day, 380K+ concurrent at peak, created at over 5,000/second — running every RL training and evaluation sandbox from V3.2 to V4.1.
The core question: as agent training scales to hundreds of thousands of concurrent agents, where do they run? Each needs an independent sandbox holding code, dependencies, test tools and services, with file/process state persisting across multi-round interactions — traditional token-generating inference can't carry that state.
Related event: DeepSeek Reveals DSec Elastic Compute Platform Behind Its RL Training(2 posts)→
More from Infra
- Starlink Now Protects Greek Trains From Collisions With 5cm-Accurate Tracking of 134 Locomotives — XFreeze · 2026-11-16
- Google donates Agent Substrate, gVisor and OpenRL to CNCF; Agent Substrate 1.0 goes GA on GKE — dl_weekly · 2026-11-16
- Nvidia scandal isn't slowing down: Huang's near-$100B power web under strain — The Verge AI · 2026-11-16
- Volume rules chips: halvarflake says Intel lost by outsourcing demand — TheKanter · 2026-11-16
- Taipei 101 lights up "Taiwan ❤️ Jensen" as tech sector rallies behind Jensen Huang — liu_mingyu · 2026-11-16
- CUDA matmul optimization walkthrough: naive kernel to 93.7% of cuBLAS — RisingSayak · 2026-11-16