Whale's May paper introduced multi-plane network architecture, influencing AI training and chip design
bookwormengr · x · 2026-09-01
This paper from May 2025 has had a profound impact on AI training network architecture and chip design.
Key Contributions:
- Introduced the concept of a multi-plane network (previously Alibaba discussed dual-plane, while DeepSeek adopted 8 planes).
- Defined the "Ideal Multi-Plane Network": A single NIC Queue Pair (QP) can simultaneously utilize all physical ports for transmission and reception, requiring native support for out-of-order placement within the NIC.
Engineering Context: DeepSeek could not fully implement this "ideal" architecture due to hardware limitations (using ConnectX-7 NICs), highlighting the gap between theoretical design and hardware reality.
More from Infra
- Saudi AI firms Humain, DataVolt partner on 100MW Red Sea data center — Polymarket · 2026-09-01
- Mixed old GPUs + 32GB RAM runs Qwen3 27B at 20t/s for local coding — pepijndevos · 2026-09-01
- ByteDance's UBASE: AI Search Engine for Trillion-Scale Vectors — _reachsumit · 2026-09-01
- LinkedIn's GPU Framework for Policy-Aligned Semantic Search — _reachsumit · 2026-09-01
- REIGN Enables Efficient Context-Length Scaling for Retrieval — _reachsumit · 2026-09-01
- Ollama Switches to Transparent Per-Token Pricing with Monthly Credit Pools — ollama · 2026-09-01