The most deployed GPU for AI training and inference. Hopper architecture delivers breakthrough performance with 80GB HBM3 memory and Transformer Engine for generative AI workloads.
Flexible pricing options to match your workload requirements.
Pay as you go with no commitment
Save 15% with monthly commitment
Maximum savings with annual commitment
Automatic mixed precision with FP8 for up to 4x throughput on transformer models.
3.35 TB/s HBM3 bandwidth eliminates data bottlenecks in training.
900 GB/s bidirectional bandwidth for seamless multi-GPU scaling.
Partition into up to 7 isolated instances for multi-tenant inference.
Train GPT-style models with billions of parameters efficiently.
Customize large language models on your proprietary data.
Serve AI models at scale with optimized Transformer Engine.
Accelerate simulations, genomics, and drug discovery workflows.