Nvidia/Mellanox engineer explains why this network architecture is a bad idea
AccBalanced · x · 2026-08-27
Someone from Nvidia/Mellanox spent 10 minutes explaining why a particular network architecture is a bad idea: higher latency, and higher cost from additional switches. See the attached illustration in the original post.
More from Infra
- Jensen prices the watt: GPU revenue per GW to hit $64B — sudoraohacker · 2026-08-27
- Colibrì Engine Update: Runs 2.8T Param Models, Boosts Speed via Expert Caching — solyarisoftware · 2026-08-27
- Jensen Huang: Data Center Investment Payback Period Under One Year — zephyr_z9 · 2026-08-27
- vLLM promotes TPU to first-class backend with new unified support — vllm_project · 2026-08-27
- 320B Model Runs on Mac: OrcaSAQ Quantization Brings GLM-5.3 to Apple Silicon — alejandroll10 · 2026-08-27
- GLM 5.3 Flash hits RunInfra day one: 254 tok/s, 1M context, $0.10/1M input — alejandroll10 · 2026-08-27