lauriewired: CXL.mem via a good switch at ~500ns should still beat 1-4us RDMA
lauriewired · x · 2026-09-07
In a debate over CXL vs RDMA (tied to the DRAM density stagnation discussion), lauriewired pushes back on the comparison: top-tier RDMA is on the order of 1-4us per request, while CXL.mem through one good switch should be 500ns. Even with a larger topology, he believes CXL stays 'firmly better than RDMA', while leaving room that the other party may have seen unusually fast RDMA.
More from Infra
- Reddit user shows NVFP4 Qwen3.8-27B matches BF16 with two sampling tweaks, runs 3x faster — UmpireBorn3719 · 2026-09-07
- Budget GPU Advice for Local LLMs: Modded 2080 Ti vs Mi50 Under $700 — Current-Set1963 · 2026-09-07
- Interview: AI's Next Big Constraint Isn't Chips — It's Energy — kimmonismus · 2026-09-07
- Fervo Energy signs 396MW geothermal deal with Google, with option for 600MW more — Beth_Kindig · 2026-09-07
- 1-bit quantized embeddings cut vector index storage up to 60x with <1% quality loss — burkov · 2026-09-07
- Starlink caps heavy 'unlimited' users to 10Mbps after ~5TB monthly usage — mcraddock · 2026-09-07