NVIDIA Benchmarks NVFP4 vs BF16 on Nemotron 3.5 Lightning, Reproducible on One H100
NVIDIA Developer · youtube · 2026-09-15
NVIDIA Developer compares NVFP4 and BF16 in practice using Nemotron 3.5 Lightning, which ships as both BF16 and NVFP4 checkpoints. The walkthrough covers memory footprint, throughput, and accuracy, showing the smaller NVFP4 checkpoint frees GPU memory with little quality loss, and includes a notebook to reproduce the full comparison on a single H100.
More from Infra
- Memphis residents say noise from Musk data-center power plant is costing them sleep — LuizaJarovsky · 2026-09-15
- SSD prices are rising as AI demand breaks decades of falling storage costs — lemire · 2026-09-15
- Analyst thesis: 2027 is peak chip constraint year, relief not until 2028 — BenBajarin · 2026-09-15
- Full-text search over all 49M HN items runs surprisingly fast on pure Neon Postgres — matei_zaharia · 2026-09-15
- Law firm Latham & Watkins buys Nvidia servers to run AI in-house as enterprises stock up on GPUs — AccBalanced · 2026-09-15
- Helion Kernels Land in Hugging Face Kernels: 1.2x Faster Than FlashAttention on H100 — RisingSayak · 2026-09-15