antirez: DeepSeek v4.1 Flash runs on a single DGX Spark at 9 t/s, 22 t/s dual via RDMA
antirez · x · 2026-09-14
antirez committed new code to DwarfStar enabling DeepSeek v4.1 Flash to run on a single NVIDIA DGX Spark with SSD streaming at 9 tokens/s generation. Dual-Spark setups over RDMA reach 22 t/s. Notable progress for running large models locally on consumer-adjacent hardware.
Related event: antirez Runs 500B-Param DeepSeek on a Single DGX Spark via SSD Streaming(3 posts)→
More from Infra
- INT21's SwarmOS runs 10,000 agents to evolve and optimize inference engines — bingxu_ · 2026-09-14
- Harvard CS249r ML Systems Vol II performance chapter covers graph compilation, fusion, CUDA profiling — blaizedsouza · 2026-09-14
- Microsoft's Brad Smith points to Quincy, WA as a model of datacenter-community coexistence — IsForAt · 2026-09-14
- ZLUDA + ROCm runs CUDA-targeted Windows apps on AMD GPUs at ~3% slower than native — _underlines_ · 2026-09-14
- Modal rebuilds sandbox platform to run 1M concurrent agent sandboxes in under a minute — charles_irl · 2026-09-14
- What if all AI API credits were condensed into a single tokenized currency? — sull · 2026-09-14