All In Data CenterAll In Data CenterCenter of the ecosystem
IT Infrastructure
DGX H200
NVIDIA By NVIDIA

DGX H200 ✓ Verified

The NVIDIA DGX H200 is a fully integrated AI supercomputing appliance with eight H200 Tensor Core GPUs interconnected by NVLink and NVSwitch, purpose-built for enterprise generative AI training and inference. It is a foundational building block of NVIDIA's AI data center platform.

Country of Origin
United States
Technology
8-GPU AI supercomputing system
Key Features
141GB HBM3e per GPUH200's larger, faster memory accelerates large-model training and inference.
Fully connected NVLinkNVSwitch fabric gives all 8 GPUs 900 GB/s of interconnect bandwidth.
Turnkey AI systemDelivered with NVIDIA AI Enterprise software stack, ready to deploy.
Technical Specifications
GPUs
8x NVIDIA H200 Tensor Core GPUs
GPU memory
1,128 GB total HBM3e (141 GB per GPU)
AI performance
32 petaFLOPS FP8
GPU interconnect
4th-gen NVLink + NVSwitch, 900 GB/s per GPU
CPUs
Dual Intel Xeon processors
Networking
NVIDIA ConnectX / BlueField high-speed fabric
Applications
Generative AI and LLM training AI inference at scale High-performance computing
Related Products
← Virtual Storage Platform 5600 GB200 NVL72 →