Gigabyte NVIDIA H100 SXM5 80GB HBM2e GPU Compute Module

The Gigabyte GV-NH100SXM5-80G is a professional GPU compute module built on NVIDIA's Hopper architecture, featuring the H100 SXM5 GPU with 80 GB of HBM2e high-bandwidth memory. Designed for integration into high-density server platforms such as NVIDIA HGX H100 baseboard configurations, this module delivers the full performance envelope of the H100 SXM5 die, including fourth-generation Tensor Cores, a second-generation Transformer Engine, and NVLink 4.0 interconnect support for multi-GPU scaling. The SXM5 form factor enables higher power delivery and thermal headroom compared to PCIe variants, sustaining a 700 W TDP to maximize sustained compute throughput across extended workloads.

At the core of this module is NVIDIA's 80-billion-parameter-class acceleration capability, with FP8 Tensor Core performance reaching up to 3,958 TFLOPS, FP16 up to 1,979 TFLOPS with sparsity, and FP64 up to 34 TFLOPS for scientific and simulation workloads. The 80 GB HBM2e memory subsystem provides 3.35 TB/s of memory bandwidth, enabling the handling of extremely large model states, datasets, and intermediate activations without off-chip bottlenecks. NVLink 4.0 supports up to 900 GB/s bidirectional bandwidth per GPU in multi-GPU configurations, making large-scale distributed training and inference across eight-GPU nodes highly efficient.

The Gigabyte GV-NH100SXM5-80G is positioned for enterprise datacenters, AI infrastructure operators, and high-performance computing facilities requiring validated, production-grade GPU compute modules. It supports NVIDIA's full software ecosystem including CUDA, cuDNN, TensorRT, and the NeMo and Triton frameworks, ensuring broad compatibility with leading AI training and inference stacks. The module is suited for bare-metal AI server builds and integration into qualified HGX platform designs serving large language model development, scientific simulation, and enterprise AI inferencing at scale.

Ideal for

• Large language model (LLM) pre-training and fine-tuning requiring high memory capacity and multi-GPU NVLink fabric connectivity
• Enterprise AI inference serving for transformer-based models using TensorRT-LLM with sustained high-throughput FP8 execution
• High-performance computing and numerical simulation workloads demanding FP64 double-precision throughput at datacenter scale
• Generative AI and multimodal model development leveraging the Transformer Engine for automated FP8/FP16 precision management
• Drug discovery and molecular dynamics simulation benefiting from high memory bandwidth and large addressable GPU memory
• Datacenter AI infrastructure build-outs within HGX H100 8-GPU baseboard configurations for consolidated, high-density GPU compute

Technical specifications

Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.