NVIDIA Announces Pascal Generation New GPU "TESLA P100" & Deep Learning Supercomputer "DGX-1"

GPU giant NVIDIA hosted "GPU Technology Conference(GTC) ", a new Pascal generation GPU"TESLA P 100Announced. Furthermore, a super computer for deep learning which carries 8 TESLA P100 "DGX-1"At the same time, clarifying the overwhelming performance that can be computed 75 times faster than the conventional system.

The NVIDIA DGX-1 Deep Learning System | NVIDIA
http://www.nvidia.com/object/deep-learning-system.html

Tesla P100 Most Advanced Data Center Accelerator | NVIDIA
http://www.nvidia.com/object/tesla-p100.html

Quadro VR Recommended Systems | NVIDIA
http://www.nvidia.com/object/quadro-vr-recommended-systems.html

NVIDIA Announces Tesla P 100 Accelerator - Pascal GP 100 Power for HPC
http://www.anandtech.com/show/10222/nvidia-announces-tesla-p100-accelerator-pascal-power-for-hpc

NVIDIA announced the new GPU series "Tesla" family. The Tesla family's most noteworthy is the high-end model "Tesla P100".


The back side looks like this. The Tesla P100 is a style pattern like a CPU LGA package that connects with two connectors.


The Tesla P100 is a GPU suitable for deep learning applications manufactured by TSMC's 16 nm FinFET process, adopting HBM 2 for memory. Half precision floating point arithmetic which is considered important in deep learning realizes 21.2 TFLOPS.


Unlike conventional GPU architecture, "NVLink" which GPU can directly access each other's VRAM is adopted, eliminating the slow data transfer speed bottleneck of CPU memory and PCI Express. By interconnecting the four GPUs with NVLink, it will realize a maximum bandwidth of 160 GB / s.


The server with Tesla P100 installed. It is scheduled to be released in the first quarter of 2017 from IBM, HP, Dell etc.


In addition, NVIDIA announced "DGX-1" super computer for deep learning that can respond to various artificial intelligence technology development, such as automatic driving car and VR simulation.


The DGX - 1 has eight Tesla P100s with 16 GB memory built into the GPU and up to 170 TFLOPS for half - precision floating - point arithmetic.


The deep learning system equipped with dual CPU Xeon E5-2697 boasts phenomenal performance that can take 150 hours to complete calculation in just 2 hours.


The DGX - 1 is scheduled to be released in 2016 with a price of 129,000 dollars (about 14 million yen).

·bonus
Unmanned race held by Formula E from 2017RoboraceAs the brain, "NVIDIA Drive PX 2NVIDIA has announced that it will be adopted.


You can check Drive PX 2 in the following article.

"NVIDIA Drive PX 2" which realizes automatic driving car with the capacity of 150 MacBook Pros with one board - GIGAZINE