NVIDIA DGX™ Systems & Platform

DGX systems are specifically designed to meet the requirements of deep learning, data science and large language models – from individual workstations right through to large-scale AI factories:

• The Blackwell generation (DGX B200 & DGX B300): Maximum performance for training, fine-tuning and inference serving of models with trillions of parameters. Equipped with state-of-the-art GPU architecture and massive HBM3e memory capacities (up to 2.3 TB on the B300), they set new benchmarks in computational and energy efficiency.
• The Hopper generation (DGX H200): The tried-and-tested foundation for demanding enterprise workloads. With 8x NVIDIA H200 GPUs (141 GB HBM3e per GPU) and integrated high-speed network interfaces, it delivers the necessary throughput for complex AI pipelines.
• Full system integration: Every DGX system is delivered as a pre-validated appliance in which computing cores, network connectivity (BlueField-3 DPU / ConnectX) and the software environment are perfectly coordinated.

Businesses need infrastructure that is ready for immediate use and delivers top performance without time-consuming compatibility testing:

• Out-of-the-box productivity: A turnkey design that minimises administrative overhead enables data science teams to get to work straight away – whether for customer experiences, process optimisation or scientific breakthroughs.
• The complete ecosystem: The combination of NVIDIA AI Enterprise, Base Command and NGC provides centralised control platforms, validated containers and SDKs that eliminate I/O bottlenecks.
• Scalable reference architectures: Seamless scalability from individual DGX nodes through DGX BasePOD™ to gigascale DGX SuperPOD™ AI factories.
• Enterprise-grade support: Direct access to specialist NVIDIA DGXperts for strategic advice and fail-safe, round-the-clock operation.