The NVIDIA HGX H100 640GB SXM5 is a high-performance AI supercomputing baseboard designed for massive-scale data center deployments. It features eight H100 Tensor Core GPUs interconnected via NVLink to provide unparalleled throughput for AI training and inference workloads.
- Architecture: NVIDIA Hopper architecture with 80 billion transistors per GPU.
- GPU Configuration: 8x NVIDIA H100 SXM5 Tensor Core GPUs.
- Memory: 640GB total HBM3 memory (80GB per GPU) with up to 3.35 TB/s bandwidth per GPU.
- Interconnect: Fourth-generation NVLink and NVSwitch providing 900 GB/s GPU-to-GPU bandwidth.
- Performance: Delivers up to 32 petaFLOPS of FP8 deep learning compute.
- Cooling: Air-cooled baseboard design optimized for standard data center environments.
- Host Interface: PCIe Gen5 support for high-speed CPU-to-GPU communication.
- Scalability: Designed for multi-node scaling in HGX clusters for exascale AI workloads.
- Transformer Engine: Specialized hardware for accelerating Transformer-based models.
Related Products
A relatable product