NVIDIA president reveals GPU development roadmap
Kepler and Maxwell — future Nvidia GPU architectures
Shortly, after hot news from IDF 2010 (Intel Developer Forum), where new Intel Sandy Bridge processors were presented, another wave of breaking news has spread round the globe. This time it is GPU Technology Conference 2010 which was held on 20th of September in San Jose, California.
NVIDIA head revealed company’s future plans. The plan focuses mainly on use of graphic solutions’ computing power in servers and supercomputers:
Fermi — it is only the beginning. Currently, there are a lot of Fermi GPUs in servers and supercomputers already. These GPUs can operate as a standard 3D graphical processor, as well as a programmable parts that can be used for visualization or other research calculations reaching 1.5GigaFlops per Watt with doubled accuracy. Performance per Watt becomes a key feature for GPU and CPU, since company tends to reduce total expenses including cooling and electricity bills which are necessary for highly productive system powering.
Later, NVIDIA president, Jen-Hsun Huang presented NVIDIA GPU development roadmap.
By 2011 new GPUs, code name Kepler, will see the light. These GPUs will be made for the 28nm process, and is supposed to have 3 to 4 times the performance/watt of Fermi. Next GPU to be released by NVIDIA is expected to appear in 2014, code name Maxwell, and it is likely 22nm-manufactured. As Jen-Hsun Huang said, Maxwell GPUs will be 16 times faster than current Fermi GPUs and performance of Maxwell will reach 16GigaFlops/Wt.