Features | H100 80GB PCIe | H100 80GB SXM | A100 80GB PCIe | A30 |
---|---|---|---|---|
FP64 | 26 TFLOPS | 34 TFLOPS | 9.7 TFLOPS | 5.2 TFLOPS |
FP64 Tensor Core | 51 TFLOPS | 67 TFLOPS | 19.5 TFLOPS | 10.3 TFLOPS |
FP32 | 51 TFLOPS | 67 TFLOPS | 19.5 TFLOPS | 10.3 TFLOPS |
Tensor Float 32 (TF32) | 756 TFLOPS | 989 TFLOPS | 156 TFLOPS | 312 TFLOPS* | 82 TFLOPS | 165 TFLOPS* |
BFLOAT16 Tensor Core | 1513 TFLOPS | 1979 TFLOPS | 312 TFLOPS | 624 TFLOPS* | 165 TFLOPS | 330 TFLOPS* |
FP16 Tensor Core | 1513 TFLOPS | 1979 TFLOPS | 312 TFLOPS | 624 TFLOPS* | 165 TFLOPS | 330 TFLOPS* |
FP8 Tensor Core | 3026 TFLOPS | 3958 TFLOPS | 624 TOPS | 1248 TFLOPS* | 330 TOPS | 661 TOPS* |
INT8 Tensor Core | 3026 TOPS | 3958 TOPS | 1,248 TOPS | 2,496 TOPS* | 661 TOPS | 1321 TOPS* |
GPU Memory | 80GB | 80GB | 80GB HBM2e | 24GB HBM2 |
GPU Memory Bandwidth | 3.35TB/s | 2TB/s | 1,935GB/s | 933GB/s |
Max Thermal Design Power (TDP) |
300-350W (configurable) | Up to 700W (configurable) | 300W | 165W |
Multi-Instance GPU | Up to 7 MIGs @ 10GB | Up to 7 MIGs @ 10GB | Up to 7 MIGs @ 10GB | 4 GPU instances @ 6GB each 2 GPU instances @ 12GB each 1 GPU instance @ 24GB |
Form Factor | PCIe | SXM | PCIe | PCIe |
Interconnect | NVLink: 600GB/s PCIe Gen5: 128GB/s |
NVLink: 900GB/s PCIe Gen5: 128GB/s |
NVIDIA® NVLink® Bridge for 2 GPUs: 600GB/s** PCIe Gen4: 64GB/s |
PCIe Gen4: 64GB/s Third-gen NVLINK: 200GB/s** |
* With sparsity
** SXM4 GPUs via HGX A100 server boards; PCIe GPUs via NVLink Bridge for up to two GPUs
Koi Computers is a 2023 NVIDIA Preferred Solution Provider. Having deployed hundreds of NVIDIA Solutions to the Federal Government, Koi Computers is a trusted and qualified partner in delivering NVIDIA solutions accelerating Artificial Intelligence, Deep Learning, and HPC Applications.
Whether you need help strategizing or designing your next data center solution, Koi Computers can help! Our expert engineering team can plan, design, validate, and size the right solution for scalability, optimal performance, and rapid efficiency.
Koi Computers’ integration team can rack and stack a complete turkey solution saving time and resources for your organization. Using only quality components and a rigorous burn-in and testing process, we ensure your data center solution is reliable.
Our engineer team can deliver fully integrated and tested rack solutions that are ready to deploy. Eliminate uncertainty with our team of professional technicians and installers, ensuring field reliability and consistency right out of the crate.
With access to our HPC Accelerate Lab and world-class Intel test tools, we can work with you to test and benchmark your application to ensure the solution meets your performance needs and lowers your total cost of ownership.