GPUs hold the promise of cheap and scalable parallel processing. But for many applications, issues with code porting, latency, and data transfer speeds can be a challlenge. Enter IceD-HPC, a new computer architecture patented by French programmer Raymond Delord.
With AMD’s Fusion, Nvidia’s Denver, Intel’s Integrated Graphics the need to make CPU closer to GPU is greater than ever. But those solutions are only targeted to Consumers for whom latency is the matter. That kind of solution wouldn’t be efficient for server, because today’s server processor market is driven by the number of cores. Think about using any transistor on a chip to make the fastest CPU, using any transistor on another chip to make the most powerful gaming GPU and unifying them with IceD to make HPC’s fastest nodes.
Download the whitepaper (PDF).