NVIDIA continues its venture for GPGPU and has announced that the successor to Fermi will launch in 2011. The GPU is code-named Kepler and will be made with 28nm technology and bring substantially better performance.
Kepler, like Fermi, is named after a famous scientist and will according to NVIDIA offer up to four times better performance per watt in 64-bit floating point double precision operations. NVIDIA did not supply any distinct launch date for the new Kepler architecture, but most likely we’re looking at a late 2011 launch.
NVIDIA has, just like AMD, been forced to postpone the transition to new nodes, but while AMD soon reveal a new graphics circuit architecture on the same node, 40nm, it looks like NVIDIA will wait with fresh launch until the more efficient 28nm node is ready and available.
Somehwat unexpected was how open NVIDIA’s VP Jen-Hsun Huang was with the company’s plans for the future during the GPU Technology Conference and not only revealed that Kepler is coming next year he also mentioned Kepler’s successor.
Maxwell is slated for launch in 2013 and once again NVIDIA has focused on GPGPU performance and parallel workloads. Maxwell will be capable of up to 12 times better performance per watt in floating point operations compared to Fermi. The exact node used for Maxwell wasn’t mentioned, but as the order goes it would seem NVIDIA is aiming for the 22nm node, which together with a more efficient architecture would offers substantial performance boosts.
NVIDIA has been working up-hill over the last year and by revealing some of its grand plans it has shown that the company is working hard to turn the tide. Jen-Hsun pointed out that hundreds of engineers are already working on Kepler and that development and manufacuting of the circuit will cost around 2 billion dollars.
How NVIDIA’s new architecture will affect gaming performance is a lot harder to predict, there was no such discussion, but we take for granted that 3D performance will not see the same great boost.
