Interconnected and Scalable
SabrePC features a range of NVIDIA HGX platforms to power the world’s most complex data centers. Each system features eight NVIDIA Blackwell GPUs with 180GB of HBM3e memory each, interconnected via NVLink for near-immediate GPU-to-GPU bandwidth at 1.8TB/s. Pair multiple HGX solutions with NVIDIA Spectrum-X and Quantum Infiniband for limitless scalability, laying the foundation for unprecedented AI and HPC performance.HGX B200 Solutions - 8x Blackwell GPUs with over 1.5TB HBM3e Memory
HGX H200 - 8x Hopper H200 GPUs with over 1.1TB HBM3e
HGX H100 - 8x Hopper H100 GPUs with 640GB HBM3
NVIDIA HGX B200, H200, & H100 Specifications
|
HGX B200 8-GPU |
HGX H200 8-GPU |
HGX H100 8-GPU |
|
|
GPUs |
8x NVIDIA B200 192GB SXM5 |
8x NVIDIA H200 141GB SXM5 |
8x NVIDIA H100 80GB SXM5 |
|
Total GPU Memory |
1.5TB HBM3e |
1.1TB HBM3 |
640GB HBM3 |
|
Aggregate GPU Bandwidth |
14.4TB/s |
7.2TB/s |
7.2TB/s |
|
NVLink |
5th Gen NVLink | 1.8TB/s |
4th Gen NVLink | 900GB/s |
4th Gen NVLink | 900GB/s |
|
NVSwitch |
4th Gen NVSwitch | 1.8TB/s |
3rd Gen NVSwitch | 900GB/s |
3rd Gen NVSwitch | 900GB/s |
|
FP64 |
320 TFLOPS |
272 TFLOPS |
268 TFLOPS |
|
FP64 Tensor |
320 TFLOPS |
540 TFLOPS |
535 TFLOPS |
|
FP32 |
640 TFLOPS |
540 TFLOPS |
535 TFLOPS |
|
TF32 Tensor |
18 PFLOPS |
8 PFLOPS |
8 PFLOPS |
|
FP16 Tensor |
36 PFLOPS |
16 PFLOPS |
16 PFLOPS |
|
FP8 Tensor |
72 PFLOPS |
32 PFLOPS |
32 PFLOPS |
|
FP4 Tensor |
144 PFLOPS |
- |
- |
Have any Questions? We got answers.
Our experts are here to help you along the way in finding the right solution for you.
