首页 > AI Servers and Accessories > Nvidia GPU > H200

NVIDIA H200 GPU

￥80000

型号：NVIDIA H200 GPU

产品特征

Memory Technology: Equipped with 141GB HBM3e memory, delivering bandwidth of up to 4.8TB/s.

4th Generation Tensor Cores: Featuring the same 4th-generation Tensor Core technology as the H100, enabling more efficient deep learning and inference.

Secure Multi-Instance GPU (MIG)NVIDIA

4th Generation NVLink, delivering up to 900GB/s of bidirectional bandwidth.

Inference Performance: Delivers up to 2x the inference performance of the H100 when running large language models such as Llama 2 70B.

商品详情
资料下载(PDF)

The NVIDIA H200 Tensor Core GPU is the latest-generation high-performance computing and

generative AI accelerator based on the Hopper architecture, purpose-built to handle complex

AI inference and high-performance computing (HPC) workloads.

Brand	NVIDIA	model	H200 141GB SXM
Item Number	H200 141GB SXM	Product dimensions	268mmX111mm double slot
Chip manufacturer	NVIDIA	Chip model	H200 141GB SXM
interface	PCIe	Memory capacity	141GB
Core position width	5120bit	graphics card slot	PCIe 5.0 x16
Video Memory Type	HBM2e/3	Memory interface	6144-bit
core frequency	1000MHz（MHz）	Memory Clock	1593MHz（MHz）
Stream processor unit	14592（个）	3D API	DirectX 12

Memory & Bandwidth

The H200 is the first GPU equipped with HBM3e memory, featuring 141GB of memory capacity and a memory bandwidth of up to 4.8 terabytes per second (TB/s), representing a 1.4x improvement over the H100’s 3.35 TB/s. This larger memory capacity and faster bandwidth significantly accelerate data processing for generative AI and high-performance computing workloads.

High-Performance Computing (HPC) Acceleration

The H200 excels in scientific computing and simulation tasks, delivering up to 110x performance gains over traditional CPUs, driven by higher memory bandwidth and an optimized compute architecture. This gives it a distinct advantage in memory-intensive applications such as scientific research and simulations.

Energy Efficiency & Cost Optimization

While delivering exceptional performance, the H200 maintains the same power envelope as the H100 (up to 700W), with higher energy efficiency and a lower total cost of ownership (TCO). This makes it not only faster but also more eco-friendly in AI factories and supercomputing systems.

Use Cases

The H200 is suitable for a wide range of scenarios including generative AI, scientific computing, and enterprise AI inference. It also supports NVIDIA AI Enterprise software, streamlining AI development and deployment workflows.