NVL72's Steep Bandwidth Cliff Inside and Outside the Rack
AccBalanced · x · 2026-07-17
The post highlights the communication disparity between the inside and outside of the NVIDIA NVL72 rack:
- GPUs inside the rack can interconnect via NVLink at approximately 1.8 TB/s;
- Once traffic leaves the rack, network card bandwidth drops to around 50 GB/s;
- This roughly 18 to 36x bandwidth gap explains a portion of NVIDIA's pricing power.
The author concludes that chips only make up half the product; the true product includes rack-level interconnects and system design.
More from Infra
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- SF Compute founder: buying compute is 'an absolutely awful experience' right now — IgorCarron · 2026-09-11
- SmolVM open-sources persistent computer infrastructure for agents that outlive chat sessions — aniketmaurya · 2026-09-11
- PyTorch Day Korea 2026 launches first offline conf, CFP closes Sept 13 — PyTorch · 2026-09-11
- Local LLM server dilemma: 4x CMP-170HX (price up 53% in 20 days) vs Mac Studio M5 Ultra — rumboll · 2026-09-11
- llama.cpp lands Flash Attention tuning for RDNA4, big prefill gains on AMD — pmttyji · 2026-09-11