Nvidia unveiled its new Pascal processor platform, a next-generation GPU that -- similar to its Maxwell, Kepler, Fermi, and Tesla GPUs -- takes its name from the 17th century polymath responsible for substantial advancements in geometry and probability theory, as well as the inventor of the mechanical calculator.
"We can now continue to scale with Moore's Law, a scale people think we'll eventually fall off of," CEO Jen-Hsun Huang said onstage at the GPU Technology Conference in San Jose, Calif., Tuesday. "But it's not going to happen with Pascal."
Nvidia's choice for Pascal represents its ambitions for its new GPU, which relies on new technologies -- 3D memory and the company's new NVLink interconnect -- to make leaps in both speed and size.
NVLink, co-developed with IBM, enables data sharing between CPUs and GPUs at speeds 5 to 12 times faster than what we have today and is meant to replace the standard PCI Express interface used in modern CPU-GPU connections. Nvidia will continue to support PCIe, but will enable NVLink for any systems that utilize Pascal.
3D memory uses a new chip-on-wafer integration that will quadruple energy efficiency while multiplying capacity by 2.5 times current standards.
Pascal will have applications in machine learning and supercomputing, as well cloud computing and gaming. Nvidia will replace the Maxwell GPU architecture -- used in this year's GeForce 800 series GPUs -- with Pascal by 2016, Huang said.