Arches Computing Systems will demonstrate their Message Passing Interface (MPI) at Supercomputing 2008. The MPI will run in both hardware and software on the Xilinx® Accelerated Computing Platform (ACP). The ACP is built around an Intel® Xeon ® 7300/7400 multi-core processor and a Xilinx Virtex®-5 FPGA-based Nallatech FSB-FPGA module. The MPI abstraction brings application portability to the world of heterogeneous embedded computing platforms comprising high-performance processors and reconfigurable logic using FPGAs. The Xilinx Accelerated Computing Platform is the first platform supported by Arches-MPI from Arches Computing Systems. Arches-MPI is available immediately from Arches Computing.
The Arches-MPI technology offers a familiar programming environment to application developers for heterogeneous embedded computing systems — such as the Xilinx ACP. Arches-MPI is compatible to the common MPI standard used for programming distributed-memory parallel applications and makes it possible for processes implemented entirely in hardware on Xilinx FPGAs to participate fully and independently as part of an MPI application. The use of an MPI abstraction layer makes scalability easy and hides the hardware details from the application layer, providing essential application portability from one heterogeneous embedded system to another.
Arches Computing Systems collaboration with Xilinx has resulted in a seamless, high-bandwidth and low-latency coupling between Xeon processors and high-performance Xilinx® Virtex®-5 FPGAs. This can be leveraged by application developers to build high-performance, heterogeneous embedded computing applications using a familiar API. An application can be developed independent of whether MPI processes are running in software on Intel® Xeon® processors or as hardware blocks in multiple Xilinx FPGAs. Programmers enjoy the same standard MPI interface across the entire platform.