Hopper-architecture accelerator with 80 GB HBM3, deployed in 8-GPU HGX training nodes.
The H100 SXM is the baseline Hopper training accelerator and remains the most widely deployed configuration for multi-GPU training clusters. It is supplied as an SXM5 module on an HGX baseboard, with NVLink providing full-bandwidth GPU-to-GPU communication inside the node.
AI trainingH100 SXM is sourced for ai training deployments within gpus, configured to the host platform and quantity specified in your requirement.
InferenceH100 SXM is sourced for inference deployments within gpus, configured to the host platform and quantity specified in your requirement.
HPCH100 SXM is sourced for hpc deployments within gpus, configured to the host platform and quantity specified in your requirement.
Supplied on HGX baseboards in 4-GPU or 8-GPU configurations. PCIe card variants are a separate part; specify which is required.
AI training · Inference · HPC
Tell us your requirement and we'll reply with sourcing options and lead time.
