A server on NVIDIA HGX B300, the Blackwell Ultra platform for training and inference of the largest models. Eight B300 GPUs with 288 GB of HBM3e each give 2.3 TB of GPU memory per system, with fifth-generation NVLink between GPUs at 1.8 TB/s. Two Intel Xeon 6700 processors with P-cores and 32 DDR5 slots (up to 4 TB at 6400 MT/s or up to 8 TB at 6000 MT/s). Eight NVIDIA ConnectX-8 SuperNIC 800 Gb/s adapters for clustering, eight hot-swap NVMe E1.S bays and 3+3 6.6 kW Titanium-class power supplies.

A proven NVIDIA HGX H200 platform for model training and inference. Eight H200 SXM5 GPUs with 141 GB of HBM3e each: more memory per GPU than the H100, so large models run without being split across nodes. Two Intel Xeon 8558 processors, 2 TB of DDR5-5600 and eight InfiniBand adapters for joining nodes into a cluster.

An 8U server on NVIDIA HGX B200 with two Intel Xeon 6700/6500 processors. Eight B200 GPUs exchange data over NVLink at 1.8 TB/s. 32 DDR5 slots (RDIMM up to 6400 MT/s, MRDIMM up to 8000 MT/s), eight hot-swap 2.5″ NVMe Gen5 bays and 12 PCIe Gen5 x16 slots for ConnectX-7 network adapters and BlueField-3 DPUs. Power: 6+6 3000 W 80 PLUS Titanium supplies.

A server built on the NVIDIA HGX B200 platform for training and inference of large language models. Eight B200 GPUs in SXM form factor are linked by fifth-generation NVLink, giving about 1.4 TB of HBM3e memory per node. Two Intel Xeon processors, up to 4 TB of DDR5 and Samsung PM9D3a NVMe drives on PCIe 5.0. For clustering: eight NVIDIA ConnectX-7 400GbE adapters and two BlueField-3 DPUs. Ships with an NVIDIA AI Enterprise license.

The infrastructure powering the generative AI revolution. Dell’s PowerEdge XE9680 is purpose-engineered for the most demanding AI workloads on the planet — training and fine-tuning large language models, running generative AI inference at scale, and supporting multi-modal AI pipelines. With 8 NVIDIA H100 or B200 Tensor Core GPUs connected via NVLink and NVSwitch, the XE9680 delivers over 64 petaFLOPS of AI compute in a single chassis. Flexible air or direct liquid cooling options mean you can deploy it in any data center environment.