On June 9, local time, AMD announced that, as the third generation of AI systems, the NVIDIA DGX A100 has become the latest high-performance computing system using the second-generation AMD EPYC processor. The flexible architecture of the NVIDIA DGX A100 delivers 5 petaflops (petaflops) of AI performance, enabling enterprises to accelerate many types of AI workloads such as data analysis, training and inference.
It is reported that the NVIDIA DGX A100 utilizes the ultra-high performance of two AMD EPYC 7742 processors, 128 cores and support for DDR4-3200MHz and PCIe 4, running at a maximum speed of 3.4 GHz. The second-generation AMD EPYC processor is the first and currently the only x86-based server processor supporting PCIe 4.0, providing high-bandwidth I/O for high-performance computing and between CPUs and other devices (such as GPUs). The connection between them is crucial.
“On a single x86 data center processor, only 2nd Gen AMD EPYC can provide up to 64 cores and 128 lanes of PCIe 4 connectivity,” said Raghu Nambiar, AMD’s global vice president of data center ecosystem and application engineering. It’s great to see the power of the NVIDIA DGX A100 system effectively doubling the I/O bandwidth. With the 2nd Gen AMD EPYC processors, our partners and customers can Maximizing performance and cost efficiency in converged infrastructure workloads gives teams the flexibility and ability to keep innovating.”
Charlie Boyle, vice president and general manager of DGX Systems at NVIDIA, said: “The NVIDIA DGX A100 system is a huge leap forward in performance and capability, and the 2nd Gen AMD EPYC processors used in the DGX A100 provide extreme performance and support fourth PCIe generation. NVIDIA has leveraged these features to create the world’s most powerful AI systems; while maintaining compatibility with the use of GPU-optimized software stacks across the DGX family.”
The Links: SP14Q002-A1 C070FW01 V0