Nvidia loves using scientists for its code names, and Pascal (named for mathematician Blaise Pascal) is the next big one coming.
Nvidia chief executive Jen-Hsun Huang announced today that its next-generation graphics chip technology, Pascal, is coming in the next year or so and will be 10 times more powerful than Maxwell, the graphics processing unit (GPU) architecture introduced in new chips for the first time last year.
Nvidia’s Maxwell-based Titan X debuted today with the ability to operate at 7 teraflops. If Pascal-based GPUs can in fact do ten times that amount, we could look forward to 70 teraflops of performance in a single chip in a year or two. Nvidia didn’t say precisely when the first Pascal chips would ship.
In a Q&A after the speech, Nvidia executive Jonah Alben said that the new Pascal architecture would use 2.5-dimensional memory stacking chip technology, and the first chip is expected in 2016. The 10X comparison refers directly to processing power related to “deep learning” apps.
Sumit Gupta, another Nvidia executive, said the chip is aimed at support 32 gigabytes of graphics memory, compared with 12 gigabytes for today’s newly announced Titan X graphics chip.