Data Center GPU
Accelerators for AI training, large-language-model inference, rendering and scientific computing.
We supply current-generation PCIe data center GPUs. H100-class accelerators are ideal for large-model training, while L40S-class cards offer a cost-effective inference and graphics+AI hybrid option.
| Training Flagship | NVIDIA H100 SXM5 / PCIe, Hopper architecture, 80 GB HBM3 |
|---|---|
| Inference / Graphics | NVIDIA L40S, Ada Lovelace architecture, 48 GB GDDR6 |
| H100 Memory Bandwidth | 3.35 TB/s |
| H100 FP8 Tensor | Up to 3,958 TFLOPS |
| L40S FP8 Tensor | Up to 1,466 TFLOPS, 350 W TDP |
| Form Factor | Dual-slot PCIe, passive heatsink for server deployment |