Petabit-Class Single-Hop Switching Paper Targets Massive MoE Training: 16x4 OSFP, 512x200G Lanes
jwt0625 · x · 2026-09-09
The tweet identifies the viral 1-petabit switch talk as the paper "Breaking Through the Network Wall: Petabit-Class Switching for Single-Hop MoE Training at Scale." The author adds hardware details: 16x4 OSFP cages and 512x200G lanes today; doubling lane rates would mean 5x OSFPs, 40 XPOs, 640 MMC connectors and 5120 active fibers with CPO—not crazy since Nvidia already does similar. The paper charts next-gen switching architectures to break the network wall for large-scale MoE training.
Related event: Marvell Teralynx T100 and Petabit Switches Aim to Break the Network Wall(2 posts)→
More from Infra
- MiniMax H3 runs fully on a Galaxy S26: 5-second video with audio in ~7 minutes — BidFunny4862 · 2026-10-27
- Unconventional AI acquires inference chip firm NeuReality in 1,000x efficiency bid — NaveenGRao · 2026-10-27
- Musk: robot RAM demand will dwarf everything else, and 100% of superintelligence compute goes to deep space — elonmusk · 2026-10-27
- RTX 5090 hits $7K as PC gaming community goes into 'denial' over AI boom — PAstynome · 2026-10-27
- Wan2GP lands on Pinokio: 1-click AI video on 6GB GPUs — cocktailpeanut · 2026-10-27
- AWS on agent prompt caching: explicit + implicit patterns to slash token costs — AWS ML Blog · 2026-10-27