Exploiting GPGPU RDMA Capabilities Overcomes Performance Limits


GPGPUs already offer performance and ease-of-use advantages for parallel processing designs. By taking advantage of RDMA features in GPGPU chips, performance is boosted further by overcoming interconnect bottlenecks.

GPU technology has increasingly found favor in military/aerospace applications because of the significant opportunity it presents to apply its inherently massively parallel architecture to non-graphics applications that can take advantage of that parallelism. However, it has not been possible to extract the maximum possible performance from so-called GPGPU—general purpose computing using graphics processing units—because of an important limitation. That limitation has now been overcome, creating the potential to extract even more throughput from the technology.


** This article appears in the April 2013 edition of COTS Journal (article begins on page 16).