Nvidia has been doing quite well for itself with its Kepler line of graphics cards. But inevitably, a new generation must come to usurp aging tech. At Nvidia's GPU Tech Conference, the company revealed their roadmap, which leads to a 2016 release of new "Pascal" cards.
Named for the 17th century French mathematician Blaise Pascal, the new generation of Nvidia GPUs promises to innovate with what the company is calling "3D Memory." By stacking DRAM chips and bringing them into the same package as the GPU, the GPU can get data from memory more quickly. The end result is several times greater bandwidth and more memory capacity.
A new feature called "NVLink" will also enable the GPU to send data at speeds faster than PCI Express. NVLink will provide between 80 and 200GB/s of bandwidth, a huge improvement over PCIe 3.0's current max speed of 8GB/s. With NVLink, the GPU will have full-bandwidth access to the CPU's memory system. "With the GPU's superior streaming ability, the GPU will sometimes be able to stream data out of the CPU's memory system even faster than the CPU," Nvidia's David Foley explained.
Pascal will also support unified memory, allowing CPUs to access GPU memory and vice versa. "Developers don't have to allocate resources between the two," Nvidia points out in their announcement. Through these improvements, Nvidia believes it can enable "a whole new class of applications."
You can see the announcement here: