During Nvidia's quarterly financial call, Nvidia's CEO, Jen-Hsun Huang, shed a bit more details regarding Nvidia's next-generation Pascal graphics processing architecture.

While there were no specific details regarding the GPU itself, Jen-Hsun Huang did reveal that Pascal GPU will be based on FinFET (fin-shaped field-effect transistor) manufacturing process, so it will be either 14nm or 16nm, depending on who will be making them. This will allow Nvidia to increase a number of CUDA cores and probably bring considerable increase in both performance and power efficiency.

In addition to the new manufacturing process, Pascal architecture will support second-generation high-bandwidth memory (HBM2). This will allow Nvidia to put up to 32GB of memory next to the GPU and bring 820GB/s to 1TB/s of memory bandwidth.

When it comes to supercomputers, Pascal chip will also integrate NVLink interconnection with 80GB/s or higher bandwidth, which will increase performance in Pascal-based Tesla accelerators for high-performance computing (HPC) applications. The same NVLink could also bring significant improvements to multi-GPU technology thanks to increase bandwidth for inter-GPU communications.

Jen-Hsun also noted that its next-generation Pascal GPU architecture will spread from Geforce to Quadro, Tesla and Tegra which is a great thing for the company but we still need to see and hear about specific details regarding the GPU itself.

Hopefully, Nvidia will share a bit more details as we draw closer to the end of this year as Pascal is expected to launch in 2016.



Source: KitGuru.com.