NVIDIA has introduced the Tesla P100 — the most-advanced data center accelerator that the company has ever built. It is based on NVIDIA's new Pascal GP100 GPU architecture that has been designed to provide higher performance and efficiency for the most demanding computing tasks.

"Our greatest scientific and technical challenges — finding cures for cancer, understanding climate change, building intelligent machines — require near-infinite amount of computing performance. We designed the Pascal GPU architecture from the ground up with innovation at every level," said Jen-Hsun Huang, CEO and co-founder of NVIDIA.

The NVIDIA Tesla P100 comes with five key features that will help today's smartest minds in developing technologies for the future.

First off, the Tesla P100 is based on the new Pascal architecture that significantly speeds up hyperscale computing and deep learning workloads. Specifically, with NVIDIA GPU Boost technology, the GPU accelerator delivers 5.3 teraflops double-precision performance, 10.6 teraflops single-precision performance, and 21.2 teraflops half-precision performance. The lightning-fast compute nodes will allow the Tesla P100 to power the fastest computers in the market.

The innovative NVIDIA NVLink also allows multiple GPUs to work together and deliver up to five times higher performance. The high-speed, high-bandwidth interconnect feature is able to transfer 160 GB per second.

The Tesla P100 also comes with HBM2 technology that delivers over three times the memory performance compared to the older NVIDIA Maxwell architecture. The new GPU memory architecture has the processor and data stacked into a single package for faster compute efficiency. It supports up to 16 GB and bandwidth can handle up to 720 GB per second. With the HBM2, there will be a significant leap in time-to-solution for heavy and data-intense tasks.

NVIDIA has also given the Tesla P100 with Unified Memory and Compute Preemption, a significantly improved programming model.

Finally, the Tesla P100 was created using the 16 nm FinFET process. With over 15.3 billion transistors, the GPU accelerator delivers higher performance and improved power efficiency.

The NVIDIA Tesla P100 GPU accelerator will be available during the start of 2017.