In an era where data is the currency of innovation, NVIDIA continues to be at the forefront of pushing the boundaries of AI computing. The NVIDIA HGX™ H200, marks a significant milestone in this journey. Based on the outstanding NVIDIA Hopper™ architecture and equipped with the revolutionary NVIDIA H200 Tensor Core GPU, this platform takes AI computing to an unprecedented level.
What makes the HGX™ H200 so special? It's the integration of HBM3e memory - a quantum leap in memory technology. With a capacity of 141 GB and a breathtaking bandwidth of 4.8 terabytes per second, the H200 sets new standards for the efficiency and speed of data processing. Compared to its predecessor, the NVIDIA A100, the H200 offers almost twice the capacity and 2.4 times the bandwidth. This means an exponential increase in performance, opening the door to even more complex AI models and high-performance computing workloads.
H200-based systems are announced for delivery in the second quarter of 2024. This platform not only promises improved performance in AI model generation, but will also revolutionize scientific computing for HPC workloads.
In our latest blog post, we take a closer look at the innovations behind the NVIDIA HGX™ H200 and explore how they are shaping the future of AI computing. Join us as we explore the next steps on the road to an intelligent future and explore the possibilities offered by the HGX™ H200.
Read press release
Learn more
Experience Next-Level Performance
Experience Next-Level Performance
Llama2 70B Inference
GPT-3 175B Inference
High-Performance Computing
Specifications
NVIDIA H200 Tensor Core GPU
Form Factor |
|---|
FP64 FP64 Tensor Core FP32 TF32 Tensor Core BFLOAT16 Tensor Core FP16 Tensor Core FP8 Tensor Core INT8 Tensor Core GPU Memory GPU Memory Bandwidth Decoders Max Thermal Design Power (TDP) Multi-Instance GPUs Form Factor Interconnect Server Options NVIDIA AI Enterprise |
H200 SXM¹ |
|---|
34 TFLOPS 67 TFLOPS 67 TFLOPS 989 TFLOPS² 1,979 TFLOPS² 1,979 TFLOPS² 3,958 TFLOPS² 3,958 TFLOPS² 141GB 4.8TB/s 7 NVDEC 7 JPEG Up to 700W (configurable) Up to 7 MIGs @16.5GB each SXM NVIDIA NVLink®: 900GB/s PCIe Gen5: 128GB/s NVIDIA HGX™ H200 partner and NVIDIA-Certified Systems™ with 4 or 8 GPUs Add-on |
2 With sparsity.