NVIDIA H100 PCIE
NVIDIA H100 PCIe Graphics Processing Unit (GPU) Module
Product Description（产品描述）
The NVIDIA H100 PCIe module is a cutting-edge data center GPU, engineered to deliver exceptional performance for AI training, inference, high-performance computing (HPC), and advanced analytics workloads. Built on NVIDIA’s Hopper architecture, it integrates powerful Tensor Cores, CUDA Cores, and high-bandwidth memory to handle the most demanding computational tasks with efficiency. Designed with a PCI Express form factor, it offers flexible integration into standard data center servers, making it ideal for enterprises, cloud service providers, and research institutions seeking to accelerate AI innovation and scientific computing.
Product Parameters（产品参数）
- Brand: NVIDIA
- Model: H100
- Form Factor: PCI Express (PCIe 5.0 x16)
- Architecture: Hopper
- CUDA Cores: 16,384 (FP32)
- Tensor Cores: 512 (4th Generation, supporting FP8/FP16/TF32/BF16)
- HBM3 Memory Capacity: 80 GB
- Memory Bandwidth: Up to 3 TB/s
- FP8 Compute Performance: Up to 672 TFLOPS
- FP16/BF16 Compute Performance: Up to 336 TFLOPS
- FP32 Compute Performance: Up to 84 TFLOPS
- DP Compute Performance: Up to 2.1 TFLOPS
- Power Consumption (TDP): 350 W / 400 W (configurable)
Specification（规格）
- Interface: PCIe 5.0 x16 (backward compatible with PCIe 4.0/3.0)
- Memory Type: High-Bandwidth Memory 3 (HBM3)
- Memory Bus Width: 5120-bit
- Operating Temperature Range: 5°C to 40°C (ambient)
- Cooling Solution: Passive or Active (server-dependent)
- Dimensions (Typical): 304.8 mm (length) x 111.15 mm (height) x 2-slot (width)
- Software Support: NVIDIA CUDA Toolkit, cuDNN, TensorRT, NVIDIA AI Enterprise
- Certifications: CE, FCC, UL, RoHS
Reviews
There are no reviews yet.