Skip to product information
1 of 1

Cisco

Cisco UCSX-GPU-H100-NVL | H100 NVL PCIe Accelerator, 94GB HBM3, 400W, 2-Slot

SKU:UCSX-GPU-H100-NVL

Stock Status: Enquire

Request Quote
Sale Sold out
Taxes included. Shipping calculated at checkout.

Description

The Cisco UCSX-GPU-H100-NVL is an NVIDIA H100 NVL Tensor Core GPU designed for large language model inference, AI training, and high-performance computing workloads in enterprise data centers. Built on the Hopper architecture with 94GB of HBM3 memory and 3.9TB/s bandwidth, it delivers exceptional performance for transformer models and generative AI applications. The 2-slot FHFL form factor integrates into Cisco UCS X-Series servers with PCIe Gen5 connectivity.

Features

Fourth-generation Tensor Cores with FP8 Transformer Engine for accelerated AI training
- 94GB HBM3 memory with 3.9TB/s bandwidth for large-scale model deployment
- PCIe Gen5 x16 interface delivering 128GB/s bidirectional bandwidth
- NVLink bridge support for multi-GPU scaling and high-speed inter-GPU communication
- Multi-Instance GPU (MIG) capability: partition into up to 7 isolated instances
- Hopper architecture with 16,896 CUDA cores and 50MB L2 cache
- Dynamic Programming (DPX) instructions for HPC acceleration
- ECC memory protection for mission-critical workloads
- NVIDIA CUDA 12.2+ and AI framework support (PyTorch, TensorFlow, JAX)
- SMBus and IPMI management interfaces for remote monitoring
- Passive cooling design optimized for enterprise server environments
- Secure Boot support and firmware management via NVIDIA tools

Warranty

All products sold by XS Network Tech include a 12-month warranty on both new and used items. Our in-house technical team thoroughly tests used hardware prior to sale to ensure enterprise-grade reliability.

All technical data should be verified on the manufacturer data sheets.

View full details

specs-tabs

Collapsible content

Technical Specifications

FAQs

Technical Specifications

GPU: NVIDIA H100 NVL Tensor Core, Hopper architecture
- Memory: 94GB HBM3, 3.9TB/s bandwidth, 6144-bit memory interface
- CUDA Cores: 16,896 shading units
- Tensor Performance: 835 TFLOPS (TF32), 1,671 TFLOPS (FP16/BFLOAT16)
- Interface: PCIe Gen5 x16
- Power Consumption: 400W TDP
- Form Factor: 2-slot Full-Height Full-Length (FHFL), passive cooling
- NVLink Support: Multi-GPU scaling with NVLink bridge connectivity
- Multi-Instance GPU (MIG): Up to 7 isolated GPU instances
- Certifications: PCIe CEM 5.1 compliant
- Operating Temperature: 0°C to 50°C ambient
- Compatible Platforms: Cisco UCS X440p PCIe Node, UCS X-Series M8 servers

FAQs

Q: What workloads is the H100 NVL optimized for?
A: The H100 NVL is engineered for large language model (LLM) inference and training, generative AI, deep learning, and HPC simulations. The 94GB memory capacity allows deployment of massive transformer models like GPT-class networks, while the 3.9TB/s bandwidth accelerates memory-bound inference tasks.

Q: How does the H100 NVL differ from standard H100 PCIe models?
A: The NVL variant features 94GB of HBM3 memory versus 80GB in standard H100 PCIe models, and delivers nearly 4TB/s memory bandwidth—roughly double the standard PCIe version. This makes it ideal for larger models that require maximum memory capacity and bandwidth for inference workloads.

Q: Can multiple H100 NVL cards be linked together?
A: Yes, the H100 NVL supports NVLink connectivity for multi-GPU configurations. NVLink bridges enable high-bandwidth GPU-to-GPU communication, allowing enterprises to scale AI workloads across multiple accelerators in a single server.

Q: What are the power and cooling requirements?
A: The card operates at 400W TDP and uses passive cooling, requiring robust server-grade airflow. It connects via PCIe 16-pin power connector (supporting 450W or 600W modes) and is designed for enterprise rack servers with adequate power delivery and thermal management, such as Cisco UCS X-Series platforms.

Q: Does this GPU support virtualization?
A: Yes, the H100 NVL supports Multi-Instance GPU (MIG) technology, allowing partitioning into up to seven independent instances for multi-tenant workloads. It also supports NVIDIA vGPU software (version 16.1 or later) and NVIDIA AI Enterprise on VMware environments.

Recently Viewed

  • Request a Quote

    Looking for competitive pricing? Submit a request, and our team will provide a tailored quote that fits your needs.

  • Contact Us Directly

    Have a question or need immediate assistance? Call us for expert advice and real-time support.

    Call us Now  
  • Contact Us Directly

    Have a question or need immediate assistance? Call us for expert advice and real-time support.

    Contact us