Supermicro H100 NVL PCIe Add-in Board 100-300000060

The Supermicro H100 NVL PCIe Add-in Board (part number 100-300000060) is built around NVIDIA's Hopper architecture, delivering cutting-edge GPU compute performance in a standard PCIe form factor suited to a wide range of enterprise server platforms. The H100 NVL variant leverages the full NVLink Bridge capability, enabling two H100 GPUs to be paired and operated as a unified, high-bandwidth compute unit with up to 600 GB/s NVLink interconnect bandwidth between cards—dramatically accelerating inter-GPU data exchange for large model training and inference workloads.

Equipped with 94 GB of HBM3 memory per GPU (188 GB total when NVLink-paired), the H100 NVL PCIe AIB provides the memory capacity required to fit exceptionally large AI models—including frontier large language models and multi-modal foundation models—entirely within GPU memory. The card supports NVIDIA's Transformer Engine with FP8 precision, enabling up to 3,958 TFLOPS of tensor-core throughput for transformer-based neural network training, alongside full BF16, FP16, TF32, and FP64 precision support for mixed scientific and AI workloads.

As a PCIe Gen 5 x16 add-in board, the Supermicro 100-300000060 integrates into standard OCP and proprietary enterprise server chassis without requiring proprietary SXM baseboard infrastructure, broadening deployment flexibility across heterogeneous datacenter environments. The card includes NVIDIA Confidential Computing support, Multi-Instance GPU (MIG) partitioning for up to seven isolated GPU instances, and full compatibility with NVIDIA AI Enterprise software, making it a production-ready platform for regulated, multi-tenant, and high-throughput AI infrastructure deployments across cloud, on-premises, and hybrid environments.

Ideal for

• Large language model (LLM) training and fine-tuning, including models exceeding 70 billion parameters, leveraging NVLink-paired HBM3 capacity and FP8 Transformer Engine throughput
• High-throughput generative AI inference serving for enterprise chatbot, code generation, and multimodal AI applications requiring low-latency, high-concurrency GPU execution
• Scientific high-performance computing (HPC) simulations including computational fluid dynamics, molecular dynamics, and climate modeling utilizing FP64 double-precision compute
• Multi-tenant AI platform deployments using NVIDIA Multi-Instance GPU (MIG) to partition a single H100 NVL into up to seven isolated, quality-of-service-guaranteed GPU instances
• Confidential AI workloads in regulated industries such as financial services and healthcare, leveraging NVIDIA Confidential Computing to protect model weights and sensitive data in use
• Retrieval-augmented generation (RAG) pipelines and vector database acceleration requiring sustained high-bandwidth memory throughput for embedding generation and semantic search at enterprise scale

Technical specifications

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