Brief description of the solution
The 16 GPU solution is centered around a GPU-to-GPU interconnect architecture via a switch chip, this system integrates PCIe 5.0 high-speed buses to deliver seamless, lossless synergy among 16 high-performance GPUs. Designed as a versatile computing engine.
- Adapted to a variety of AI accelerator cards
it unifies AI training and inference, supports multiple AI acceleration cards, and precisely targets varied AI workloads.
- lTopological architecture
flexible topology and smart resource orchestration
- Supports multiple topologies
Resources can be intelligently allocated according to different application scenarios,Avoid wasting computing power and drive efficient operation of computing power
Technical Specifications of the Solution
- 01
2 Broadcom cards PCIe Switch expansion board
A solution featuring two Broadcom 89144 PCIe Switch expansion boards for 16-GPU interconnection
- 02
Bus Standard
PCIe 5.0,PCIe 4.0 high speed interface
- 03
PCIe Switch Solution
16GPU fully interconnected, non-blocking architecture
