NVIDIA co-designs its GPUs and models together — training in 4-bit NVFP4 from the start because Blackwell was built for fast 4-bit math — and the real bottleneck for better models is not data or compute but the diversity of reinforcement learning training environments.

NVIDIA trained Nemotron in 4-bit precision from the first step because Blackwell GPUs were designed for fast 4-bit math, avoiding the accuracy loss of post-training quantization. For capability, the team scales RL across over a million rollouts, and Bryan identifies environment diversity — not data or compute — as the true bottleneck. ✦ AI generated

Bryan Catanzaro · ByteByteGo Newsletter · 2026-07-27 · original ↗

Its larger models are pretrained in a 4-bit number format called NVFP4, meaning most of the math during training uses just four bits per value. Fewer bits means less memory and less data to move, so the math runs faster and on less power. Note that four bits is very little precision. Most teams train in higher precision and only shrink the model afterward, which loses accuracy. NVIDIA trained in 4 bits from the first step because it knew its next GPU generation, Blackwell, was being built with fast 4-bit hardware. The model and the chip were designed for each other. The first step to inventing something amazing is believing that you can. With those efficiency choices in place, post-training is where the model gains its capability. Scaling up this reinforcement learning stage has been shown to improve a model's capability significantly. Because its models are cheaper to run, it can scale RL at a lower cost, letting the model practice across many environments in parallel and learn from over a million rollouts. This is where the VP placed the real bottleneck for better models. Not data or compute. The diversity of these training environments.

Read full article ↗excerpt · fair-use quotation