NVIDIA quietly cut Vera Rubin's HBM4 bandwidth 13% per GPU — the spec that matters for inference
AIFlow_ML · x · 2026-10-28
AIFlowML analyzes NVIDIA's quiet spec cut for Vera Rubin NVL72, triggered by Jukan's Oct 26 post (700K views in a day) that HBM4 pin speed was officially lowered. Versus GTC 2026/archived pages: HBM4 bandwidth per GPU drops from 22 to 19.2 TB/s (-13%), NVLink 3.6 to 3 TB/s (-17%), rack HBM4 1580 to 1400 TB/s (-11%), rack NVLink 260 to 216 TB/s (-17%), HGX Rubin NV8 HBM4 down 26%. Capacity and compute are unchanged (288 GB, 50 PFLOPS) — but bandwidth is the speed limit for inference, making this the spec that matters.
More from Infra
- Qwen3.8 Flash Next at IQ3_S hits 275 tok/s on a single RTX 5080 via Basalt fork — yehyakar · 2026-10-29
- Google Cloud CEO: Apple Scaling Foundation Model Pretraining on Latest TPUs — shayan_ · 2026-10-29
- FlashAttention vs standard attention in one chart: the bottleneck is data movement, not compute — cneuralnetwork · 2026-10-29
- Particle-1 inference goes live on OpenRouter, claiming 28-52% fewer output tokens — ycombinator · 2026-10-29
- Qwen3.8-27B EXL3 quant hits lowest KLD of 0.0002 in just 23 GB — suprjami · 2026-10-29
- MLX-Serve 26.11.1 rewrite: plugin architecture adds models and quants without recompiling — TheMoonMidas · 2026-10-29