fal tests Serverless cluster mode for on-demand 1K H200 RDMA clusters
isidentical · x · 2026-09-21
fal's isidentical says the new cluster mode in fal Serverless can scale inference and post-training across a single 1K H200 RDMA domain — effectively on-demand access to large clusters. fal Serverless already powers 1,300+ production endpoints (895 GPUs in use, 190 req/s), and early access is open via DM.
More from Infra
- One RTX 3090 ran Qwen 27B autonomously for 3 weeks — it shipped working CUDA kernels — skeole · 2026-09-21
- Intel CEO says company can only meet ~50% of server CPU demand — Beth_Kindig · 2026-09-21
- ExLlamaV3 3bpw local Flash benchmarks: 1500 tps prefill on a single RTX 5090 — youcloudsofdoom · 2026-09-21
- SGLang patch brings MCP support: computer use and browser use agents now work — TheZachMueller · 2026-09-21
- Huawei unveils Ascend 960 supernode: 4,096 cards, 8 EFLOPS FP8, NPO optics, Q3 2027 — teortaxesTex · 2026-09-21
- Chinese models' token share on Vercel grew ~5x since June 30 — menhguin · 2026-09-21