NVIDIA's future GPU is still being manufactured by TSMC

At the GTC 2010 GPU Technology Conference, Huang Renxun announced the code number and simple case of the next two generations of GPU architectures. The foundry situation has gradually become clear.

After Fermi, the next stop of the NVIDIA GPU is called Kepler and then Maxwell, both famous physicists. The manufacturing process is upgraded to 28nm, respectively. 20nm. According to Huang Renxun, Kepler will improve double-precision floating-point performance at least four times, while introducing virtual memory, non-blocking I / O and other similar CPU functions, Maxwell will increase the performance per watt double-precision floating-point At least ten times that of Fermi.

As we all know, TSMC, GlobalFoundries are actively deploying 28nm technology, and will be launched in the future 20nm, especially after the latter separated from AMD has been trying to expand the scope of cooperation, will gradually start OEM AMD's GPU, is also actively displayed recently 28nm process development. So, will GlobalFoundries and NVIDIA come together?

The latest news shows that Kepler and Maxwell will still hand over to TSMC for OEM and will not be transferred to GlobalFoundries. Although the immatureness of the 40nm process has caused a lot of trouble, it seems that NVIDIA is still confident of TSMC and has continued for many years. Cooperation will continue.

According to the roadmap published by Huang Renxun, Kepler will be released in 2011 and Maxwell will debut in 2013, basically in line with the progress of TSMC's technology.

In addition, there is news today that AMD's next-generation performance-level GPU core "Barts" has progressed very smoothly. Each graphics card vendor is already actively preparing new cards. It will be released in October, with the PRO version (6750?) winning the GeForce. GTX 460, XT version (6770?) is easily beyond Radeon HD 5850.