From X-bit Labs: Rambus, a leading developer of memory and interface technologies, this week demonstrated an XDR-based memory sub-system that ran at 7.20GHz. Rambus claims that despite of very high clock-speed, the XDR-powered system consumes 40% less power compared to GDDR5 memory sub-system. “Future graphics and multi-core processors require significantly higher memory performance under extremely challenging power and thermal constraints. This technology demonstration highlights the outstanding power efficiency of the XDR and XDR2 memory architectures at performance levels from 3.2 to 7.2Gbps with scalability to well over 10Gbps,” said Martin Scott, senior vice president of research and technology development at Rambus. This demonstration featured Elpida's recently-announced 7.20GHz x32 1Gb XDR DRAM device and an XIO memory controller transmitting “realistic” data patterns. Rambus has not revealed the actual bandwidth provided by the XDR-based memory sub-system, but indicated that the XIO memory controller is up to 3.5 times more power efficient than a GDDR5 controller, and the total memory system can provide is up to two times more bandwidth than GDDR5 at equivalent power. At present, graphics cards manufacturers utilize up to 3.90GHz GDDR5 memory, which gives graphics chips 124.80GB/s of memory bandwidth in case of 256-bit memory controller. In case Rambus’ demonstration included 128-bit memory controller, the peak bandwidth would be 115.2GB/s, but in case of 256-bit memory controller the bandwidth would be whopping 230.40GB/s. It should be noted that Samsung Electronics began to produce 7.00GHz GDDR5 memory chips, which puts the industry-standard GDDR5 just a little behind proprietary XDR in terms of peak performance. In addition, the XIO memory controller demonstrated bi-modal operation with support for both XDR DRAM as well as next-generation XDR2 DRAM. View: Article @ Source Site |