Blackwell succeeded Hopper as NVIDIA's AI compute generation: second-gen transformerA neural network architecture using attention — the foundation of modern large language models. engines, fifth-gen NVLink, and GB200 systems pairing two B200 GPUs with a Grace CPUCentral processing unit — the primary chip that executes program instructions.. Jensen Huang called it a processor for the generative AIModels that create new content — text, images, audio, or code — from learned patterns. era — the hardware layer behind the next wave of frontier model training and inferenceRunning a trained model to produce predictions — as opposed to the training phase that learns weights..

Blackwell defined the GPUGraphics processing unit — a chip optimized for parallel math; now used for graphics, ML, and scientific computing. generation that trained and served 2024–2026 frontier models.

It raised the ceiling on AI factory scale, efficiency, and real-time trillion-parameter inferenceRunning a trained model to produce predictions — as opposed to the training phase that learns weights..