HPE Nvidia H200 NVL 141GB HBM3e PCI-Express 5.0 x16 Graphics Card - P72484-001

The HPE NVIDIA H200 NVL (P72484-001) is an HPE-certified variant of NVIDIA's flagship H200 NVL data center GPU accelerator, validated for HPE ProLiant server platforms. The P72484-001 is a distinct HPE SKU from the P72159-001 — verify compatibility with your specific HPE server configuration before ordering. Featuring 141GB of HBM3e memory, 4.8 TB/s memory bandwidth, and NVIDIA's Hopper architecture with PCIe 5.0 x16 interface, it delivers exceptional performance for AI training, large language model (LLM) inference, and high-performance computing (HPC).

Key Specifications

• Part Number / MPN: P72484-001
• Manufacturer: HPE (NVIDIA OEM)
• Product Line: NVIDIA H200 NVL
• Architecture: Hopper (GH100)
• Memory: 141GB HBM3e
• Memory Bandwidth: 4.8 TB/s
• Interface: PCI-Express 5.0 x16
• Form Factor: PCIe (NVL)
• TDP: 700W
• Cooling: Passive (requires server airflow)
• OEM: HPE-certified variant

Compute Performance

• FP8 Tensor Core: Up to 3,958 TFLOPS
• FP16 / BF16 Tensor Core: Up to 1,979 TFLOPS
• FP32: Up to 67 TFLOPS
• TF32 Tensor Core: Up to 989 TFLOPS
• INT8 Tensor Core: Up to 3,958 TOPS
• Memory Bandwidth: 4.8 TB/s (HBM3e)

Memory

• Capacity: 141GB HBM3e
• Bandwidth: 4.8 TB/s
• ECC: Yes — hardware ECC for data integrity

Features

• 141GB HBM3e — largest GPU memory for fitting massive LLMs in a single GPU
• NVIDIA Hopper architecture with 4th-gen Tensor Cores and Transformer Engine
• FP8 precision for accelerated AI inference and training
• HPE-certified firmware and iLO integration for ProLiant server management
• Passive cooling — designed for high-airflow data center environments
• Supports CUDA, cuDNN, TensorRT, NCCL, and NVIDIA AI Enterprise software stack
• MIG (Multi-Instance GPU) support for workload partitioning

Physical & Environmental

• Form Factor: Full-height, full-length (FHFL) PCIe card
• TDP: 700W
• Cooling: Passive (server airflow required)
• Operating Temperature: 0°C to 35°C (inlet air)

Compatibility

• HPE-certified for HPE ProLiant Gen11 platforms — verify specific server model compatibility with HPE QuickSpecs
• Compatible with PCIe 5.0 x16 and PCIe 4.0 x16 server slots
• Requires 700W slot/system power budget per GPU
• Managed via HPE iLO for health monitoring and firmware updates
• Compatible with NVIDIA AI Enterprise, CUDA 12.x, PyTorch, TensorFlow, JAX
• Ideal for LLM training and inference, HPC simulation, and scientific computing