NVIDIA H100 PCIE
-
Product NameNVIDIA H100 PCIe Data Center GPU Module
-
Product DescriptionThe NVIDIA H100 PCIe is a high-performance data center GPU module based on the Hopper architecture, designed for PCI Express 5.0 x16 slots. It targets compute-intensive workloads such as artificial intelligence training and inference, high-performance computing (HPC), scientific simulation, and large-scale data analytics. Unlike the SXM form factor H100 which is optimized for multi-GPU server nodes with NVLink Switch Systems, the PCIe variant offers flexible deployment in standard PCIe-based servers, enabling seamless integration into existing data center infrastructures without requiring specialized NVLink-compatible motherboards. It supports advanced technologies including Tensor Cores (with FP8, BF16, FP16, FP32, and FP64 precision), Transformer Engine for accelerating large language models (LLMs), and PCIe 5.0 for high-bandwidth data transfer between the GPU and CPU.
-
Product Parameters| Parameter Categories | Specific Parameters || —- | —- || Architecture & Core | Hopper architecture; 80 SMs (Streaming Multiprocessors); 132,710,400 transistors; TSMC 4N process || Compute Capability | Compute Capability 9.0 || Tensor Core Performance | FP8: Up to 1,979 TFLOPS; BF16/FP16: Up to 989 TFLOPS; FP64: Up to 62 TFLOPS; FP32: Up to 31 TFLOPS || Memory | HBM3 memory; Capacity: 80GB; Memory Bandwidth: 3,350 GB/s; Memory Bus Width: 5,120-bit || PCIe Interface | PCIe 5.0 x16; Peak Bandwidth: 64 GB/s (bidirectional) || Power Consumption | Typical TDP: 350W; Max Power: 400W || Form Factor Compatibility | Full-height, full-length (FHFL); Dual-slot or triple-slot cooling design || Software Support | NVIDIA CUDA, cuDNN, TensorRT, NVIDIA AI Enterprise Suite; Compatible with Linux and Windows Server operating systems |
-
Specifications| Specification Items | Details || —- | —- || Physical Specifications | Dimensions: 312mm (length) × 112mm (height) × 42mm (depth, dual-slot); Weight: Approximately 1.2kg || Cooling Requirements | Active cooling via PCIe slot fan or server chassis cooling system; Operating temperature range: 0°C to 85°C (GPU die) |