This week at the SC20 virtual conference, Xilinx will display a technology demonstration which shows the integration of Xilinx Alveo accelerator card and AMD ROCm open software platform. The technology preview is based on AMD's leading position in high-performance computing technology, especially the use of user-mode queuing and shared virtual memory, which can could distribute work directly with low latency on the Alveo
Alveo accelerators are used to provide computing, network and storage acceleration in high-performance computing applications. Xilinx devices play a vital role in accelerating infrastructure and dedicated computing, working with GPUs and CPUs to support many of the world's most demanding workloads, such as machine learning inference, real-time video transcoding, and database analysis.
These new hardware functions realize the in-depth and extensive integration of Xilinx FPGA and AMD ROCm open software platform, laying the foundation for the seamless integration of AMD Instinct GPU accelerator and our Alveo accelerator in computing, network and storage solutions.
Technology demonstration
- Use the fusion runtime in the AMD ROCm open software platform to uniformly discover and retain AMD and Xilinx accelerators
- Use the same user space queue as the low-latency work assigned to the AMD Instinct accelerator to assign the work to the Alveo accelerator
- Use the common shared virtual address space to access memory on GPU, CPU and FPGA devices