RDMA transfers burn dozens of cores: lessons on CPU QoS and high-density memory
sloppenheimer · x · 2026-09-24
The author recounts running 32-core pods just to move data over the network at RDMA/ultra-ethernet speeds — transfers consume far more CPU than expected. He wishes for ways to bypass userspace and is eyeing high-density memory and CPU QoS handling next.
More from Infra
- Zilliz CTO: agents make the enterprise data layer impossible to ignore — No_Engineer_1224 · 2026-09-24
- Running Android emulator + Chrome with 60fps streaming in a $0.072/hr cloud VM for always-on agents — cem2ran · 2026-09-24
- NVIDIA's DGX Spark Appears Unavailable, May Never Return to Sale — GabGarrett · 2026-09-24
- Not every AI task needs an LLM: 'decide' may become a standard model call — bigdata · 2026-09-24
- Prime Intellect launches Prime Sandboxes: MicroVM sandboxes purpose-built for RL training — xeophon · 2026-09-24
- M5 Ultra 96GB local inference: 3,200 tok/s aggregate prefill over 112M tokens with custom MLX server — Every-Fortune-3151 · 2026-09-24