The tremendous processing horsepower in Xilinx FPGAs has enabled its increased usage in Accelerator and Offload Engine applications. This coupled with their traditional strength in chip interfacing, given our flexible serial and parallel IO technology, allows Xilinx to support a broad range of acceleration architectures, said Amit Dhir, senior director of Aerospace, Defense and High-Performance Computing at Xilinx. By working with Intel to deliver a solution that supports the Intel(R) QPI protocol, we can simultaneously leverage the FPGA processing power and the improved bandwidth, latency and coherency capabilities of Intel(R) QPI to enable more advanced accelerator subsystems in next generation HPC systems that are based on Xilinx FPGAs and Intel(R) Xeon(R) processors.
Xilinx FPGAs support a multitude of high speed parallel and serial connectivity standards and when applied as co-processors can significantly accelerate high performance scientific, oil and gas, financial, aerospace and defense, communications, networking, and life sciences computing applications. Utilizing Intel(R) server platforms and Intel(R) QPI gives customers the highest performance and flexibility to accelerate their HPC application workloads.
Using Intel(R) QPI in-socket accelerators will be able to provide the additional capability of keeping the data coherent with the processor memory, allowing many applications to benefit from new features like reduced latency to system memory, cache coherency, shared memory models and faster/wider serial interconnect to the Intel(R) Xeon(R) processors, said Allan Cantle, president and founder of Nallatech. Xilinx’s support for this technology means customers can take advantage of the features of Intel(R) QPI technology in high performance computing applications where FPGAs can significantly improve performance per watt per dollar.