Nvidia prepping Kepler in 2011, Maxwell in 2013

— 1:22 PM on September 21, 2010

GTC -- During his opening keynote address at Nvidia's GPU Technology Conference today, Nvidia CEO Jen-Hsun Huang revealed the first details of his firm's future GPUs.


Huang offered a brief roadmap sketch, with a few notable details about a couple of future products.  He projected performance using an unusual, GPU-computing-focused metric: double-precision gigaFLOPS per watt, which indicates power-efficiency rather than raw, peak performance.

The next major GPU architecture from Nvidia, code-named Kepler, is slated for release in 2011, with the first products likely coming in the second half of the year.  Kepler will be based on a 28-nanometer fabrication process, and Nvidia intends for it to be nearly three times more efficient, in DP FLOPS per watt, than today's Fermi architecture.  Huang noted that such an improvement goes "far beyond" what process technology advances alone can achieve.  Changes in the chip architecture, design, and software will contribute to that advance, as well.

The Maxwell architecture will come next, in 2013, and will be produced on a 22-nm fabrication process.  Maxwell promises nearly an 8X increase in DP FLOPS per watt beyond today's Fermi chips.  Huang noted that, in parallel computing, power is the primary constraint, which is why he chose that metric to describe future architectures.

Between now and the Maxwell generation, Nvidia plans to introduce several compute-focused features into its GPU architectures, including pre-emption.

In between these major architectural generations, which are planned to track with the introduction of new process technologies, Huang said Nvidia will continue to produce a "mid-life kicker" or an enhancement to its existing microarchitecture. The next such update will be a refresh of the Fermi-generation GPUs, coming next year.

As before, the firm aims to do a top-to-bottom refresh of its GPU lineup, whenever it introduces a new generation of GPUs, within about three months.  Huang admitted the Fermi generation has taken longer than he'd like because of some challenges in producing the first chips.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.