Technical Specifications
architecture
NVIDIA Hopper
platforms
H200, H100, H100 NVL, HGX H200, GH200 Grace Hopper
tensor cores
4th Generation
transformer engine
FP8
memory
HBM3 / HBM3e
memory capacity
Up to 141GB
memory bandwidth
Up to 4.8 TB/s
interconnect
NVLink 4.0, NVSwitch
cuda support
1
tensorRT
1
cudnn
1
gpudirect storage
1
nvidia ai enterprise
1
frameworks
CUDA, PyTorch, TensorFlow, JAX, TensorRT, RAPIDS
primary workloads
Large Language Models, Generative AI, AI Training, AI Inference, Machine Learning, Deep Learning, High Performance Computing, Scientific Computing, Computer Vision, Recommendation Systems
deployment
Enterprise AI, AI Factories, Cloud GPU, Data Centers, Research Labs, Supercomputers
Overview
The NVIDIA H-Series GPU family is built on the revolutionary NVIDIA Hopper architecture and is designed to accelerate the world's most demanding artificial intelligence, machine learning, large language model (LLM), scientific computing, and high-performance computing (HPC) workloads. These enterprise-class GPUs deliver exceptional compute performance, advanced Tensor Core technology, ultra-fast HBM3 and HBM3e memory, and high-speed NVLink connectivity for modern AI infrastructure.
The Hopper architecture introduced major advancements over previous GPU generations, including the Transformer Engine with FP8 precision, fourth-generation Tensor Cores, confidential computing capabilities, advanced multi-instance GPU (MIG) support, and significantly improved memory bandwidth. These innovations enable enterprises to train larger AI models, deploy real-time inference services, and execute complex scientific simulations with greater efficiency.
The H-Series product lineup includes the NVIDIA H100 Tensor Core GPU, NVIDIA H100 NVL, NVIDIA H200 Tensor Core GPU, HGX H200 platform, and GH200 Grace Hopper Superchip. Each platform is optimized for different AI deployment scenarios, ranging from enterprise inference clusters to hyperscale AI training systems and next-generation supercomputers.
Organizations worldwide rely on NVIDIA Hopper GPUs for generative AI, ChatGPT-style applications, retrieval augmented generation (RAG), recommendation engines, autonomous driving, healthcare AI, financial modeling, digital twins, robotics, cybersecurity, computer vision, weather forecasting, pharmaceutical research, and scientific simulations.
With support for CUDA, TensorRT, cuDNN, NVIDIA AI Enterprise, NVLink, NVSwitch, GPUDirect Storage, and leading AI frameworks such as PyTorch, TensorFlow, JAX, and RAPIDS, the H-Series seamlessly integrates into enterprise AI environments while delivering industry-leading performance and scalability.
Whether deploying private AI infrastructure, GPU cloud services, enterprise inference clusters, or large-scale supercomputers, NVIDIA H-Series GPUs provide a proven platform for accelerating AI innovation and high-performance computing across every industry.
Popular Platforms
NVIDIA H200 Tensor Core GPU
NVIDIA H100 Tensor Core GPU
NVIDIA H100 NVL
HGX H200
GH200 Grace Hopper Superchip
Ideal Workloads
Large Language Models (LLMs)
Generative AI
AI Training
AI Inference
Machine Learning
Deep Learning
High Performance Computing
Scientific Computing
Computer Vision
Recommendation Systems
Data Analytics
Digital Twins
Drug Discovery
Financial Modeling
Robotics
Autonomous Vehicles
Healthcare AI
Cloud GPU Infrastructure
Enterprise AI
Supercomputing
Frequently Asked Questions
What are NVIDIA H-Series GPUs used for?
NVIDIA H-Series GPUs are designed for AI training, LLM inference, generative AI, machine learning, high-performance computing, scientific research, and enterprise-scale accelerated computing.
Which GPUs are included in the NVIDIA H-Series?
The H-Series includes NVIDIA H100, H100 NVL, H200 Tensor Core GPU, HGX H200, and the GH200 Grace Hopper Superchip.
What is Hopper architecture?
Hopper is NVIDIA's enterprise GPU architecture featuring fourth-generation Tensor Cores, Transformer Engine, FP8 precision, NVLink, and HBM3/HBM3e memory, delivering major improvements in AI training and inference
Are H-Series GPUs suitable for LLMs?
Yes. NVIDIA H-Series GPUs are widely deployed for training and inference of large language models, generative AI applications, and enterprise AI workloads.
