H100 SXM GPU

The NVIDIA H100 SXM is a datacenter GPU based on the Hopper architecture, representing NVIDIA's previous-generation high-performance computing and AI accelerator before the current GB300 series. This server-class GPU delivers substantial computational power for large-scale AI training and inference workloads, featuring 80GB of HBM3 memory and 14,592 CUDA cores paired with 456 4th Generation Tensor Cores. Key technical specifications include 3,350 GB/s memory bandwidth, 990 TFLOPS of FP16 performance, and support for FP8 precision through the Transformer Engine. The H100 SXM incorporates Multi-Instance GPU (MIG) technology for workload isolation and includes NVIDIA Confidential Computing capabilities for secure processing. With a 700W TDP, this GPU requires robust cooling infrastructure typical of enterprise server deployments. In cloud environments, the H100 SXM serves demanding AI research, large language model training, and high-performance computing applications where substantial memory capacity and computational throughput are essential. The SXM form factor enables high-density server configurations with NVLink interconnect for multi-GPU scaling.