NVIDIA HGX H100 8-GPU delivers 268 teraFLOPS of FP64 performance, over 3x last generation NVIDIA HGX A100.
The NVIDIA HGX H100 utilizes Fourth Generation NVLink for up to 900GB/ of GPU-to-GPU interconnect, almost 10X higher than PCIe Gen4.
NVIDIA NVSwitch™ creates a unified networking fabric that allows the entire node of multiple HGX to function as a single gigantic GPU.
Build your ideal system
| HGX H100 4-GPU | HGX H100 8-GPU | |
|---|---|---|
| GPUs | 4x NVIDIA H100 80GB SXM5 | 8x NVIDIA H100 80GB SXM5 |
| GPU Memory | 320GB HBM3 | 640GB HBM3 |
| Per GPU Memory Bandwidth | 3.35TB/s | 3.35TB/s |
| Aggregate Memory Bandwidth | 13TB/s | 27TB/s |
| NVLink Generation | Speed | 4th Gen NVLink | 900GB/s | 4th Gen NVLink | 900GB/s |
| NVSwitch Generation | Speed | N/A | 3rd Gen NVSwitch | 900GB/s |
| FP64 | 134 TFLOPS | 268 TFLOPS |
| FP64 Tensor Core | 268 TFLOPS | 535 TFLOPS |
| FP32 | 268 TFLOPS | 535 TFLOPS |
| TF32 Tensor Core | 3958 TFLOPS | 7915 TFLOPS |
| FP16 Tensor Core | 7915 TFLOPS | 15830 TFLOPS |
| FP8 Tensor Core | 15830 TFLOPS | 31662 TFLOPS |
To unlock next-generation discoveries, scientists look to high performance computing to understand complex molecules for drug discovery, simulation of mechanical and fluid dynamics, parse through length genome sequences, extrapolate data science research, and train next generation generative AI.
H100 continues to deliver high performance innovation, tripling the FLOPS in double precision FP64 Tensor Cores over the A100. AI-fused HPC applications can also leverage H100’s TF32 precision to achieve one petaflop of throughput for single-precision matrix-multiply operations, with zero code changes.
NVIDIA AI Enterprise enables users to harness the power of AI through an optimized and streamlined development and deployment framework. Coupled with NVIDIA Enterprise Support and Training Services, developers can leverage a professional to assist and teach the best AI practices. Train and deploy the best AI model, tailored for your deep learning goals today.
