GLM-5.3-Flash hits 45 tok/s on 4x DGX Spark with switchless RoCE
AccBalanced · x · 2026-09-01
Alex Ellis released a reproducible recipe for serving GLM-5.3-Flash (NVFP4) across 4x NVIDIA DGX Spark nodes via a switchless RoCE ring and DFlash2 speculative drafter. The setup achieves Tensor Parallelism 4 (TP4), offering 45 tok/s on real agentic traffic with a 262K context window. The architecture is model-agnostic and has also been tested with GLM-5.2 and DeepSeek-V4-Flash.
More from Infra
- RL train-sample systems are frontier AI's hardest infra problem — and a trade secret — brianryhuang · 2026-10-16
- Local tuning of GLM-5.3-Flash on 2×GB10 lifts decode from 49 to 58 tok/s — EAccelerate_42 · 2026-10-16
- 3x DGX Spark cluster runs GLM-5.3-Flash locally behind a proxy, 1.75M tokens/day at 119W — Teknium · 2026-10-16
- Optuna creator Akiba: when 2 hyperparameters cost $20M, what's left for BO/HPO — Stefania_druga · 2026-10-16
- Amazon Proposes Per-Request Validated Evidence Caching for Production RAG — _reachsumit · 2026-10-16
- LinkedIn's TransX report: sequence transduction for recsys, +6.02% CTR in A/B tests — _reachsumit · 2026-10-16