KEPLER AND MAXWELL were the codenames of Nvidia's future processor generations made public today at the company's GPU Technology conference keynote speech given by its CEO Jen-Hsun Huang.
After spending almost two hours talking about GPU technologies applications to super computing Huang announced that he would reveal the codenames of the company's future processor generations. They are Kepler and Maxwell, named after German astronomer Johannes Kepler and Scots mathematician James Clark Maxwell the father of classical electromagnetic theory.
Huang described the Kepler generation as being a 28nm scale design and having three to four times the performance per Watt of the company's existing Fermi generation. Kepler will go into production "later next year" and Maxwell will arrive in 2013.
A crude chart showed Fermi to have a performance of about 2 gigaFLOPS per Watt. Kepler could be seen on the chart to have about a 5 gigaFLOPS per Watt. While Maxwell had a whopping 15 or 16 gigaFLOPS per Watt. Huang stated that Maxwell would have 16X the performance of Nvidia's last generation Tesla.
"We believe we can stay on this ramp [of performance growth]," Huang told the GPU Technology audience. He added that hundreds of engineers were working on Kepler and by the time it is finished the company would have invested "a couple" of billion dollars.
According to Huang chip design will not be the only way of achieving the performance he expects. He foresees the use of virtual memory and pre-emption and autonomous processing for less blocking. µ