Dual PCIe-switch expansion board explained: 8 GPUs off one x16 slot
TheZachMueller · x · 2026-10-10
Zach Mueller (Hugging Face) breaks down how a GPU expansion board works, apologizing for earlier misreadings:
- The board carries two PCIe switches, so a single motherboard x16 slot can drive 8 (or two) GPUs
- Two topologies for the main GPU slots: direct connect via two x16-to-4x x8 MCIO cables, or cascading via a single x16-to-2x x8 MCIO cable
- The difference comes down to PCIe lanes/bandwidth per GPU — practically relevant for planning multi-GPU inference or training rigs
More from Infra
- Cascade GPU Topology Can Be Slower: The PCIe Hop Trap in Multi-GPU P2P — TheZachMueller · 2026-10-10
- Custom CUDA Megakernel Hits 140 tok/s on Qwen3.8-27B With a Single RTX 3090 — Adorable_Weakness_39 · 2026-10-10
- Keeping the cascade GPU setup requires another direct plus a retimer — TheZachMueller · 2026-10-10
- HK GPU market check: 5090 rigs near HK$90k, 5090 discontinuation rumor gains credence — karminski3 · 2026-10-10
- Orbio raises $1.2M to build rails for agents to provision their own compute — econoar · 2026-10-10
- AI value shifts from model layer to data center and apps as harness beats models, analyst argues — RihardJarc · 2026-10-10