NVIDIA Corporation (NASDAQ: NVDA) unveiled a new Pascal GPU called Tesla P100 that the company claims is ideal for scientific calculations and deep learning.
The Tesla P100 is plugged into supercomputers through PCI-ExpRess ports. The Pascal GPU will be used to hasten processes such as weather modeling and economic forecasting. The GPU can also be installed in servers and embedded into autonomous vehicles, drones, and robots to help in detecting objects. Further, it will be used in data centers that use deep learning systems to enhance speech recognition, natural language processing, classification and cloud-based image identification. One Nvidia’s DGX-1 system with eight Tesla P100 GPUs will cost $129,000.
Nvidia says that some server manufacturers, such as Hewlett Packard Enterprise Co (NYSE: HPE), Dell Inc. (NASDAQ: DELL) and Cray Inc. (NASDAQ: CRAY), should start receiving deliveries for their orders in the fourth quarter of 2016. Nvidia had introduced a version of the Tesla P100 during its GPU conference two months ago, but that version was for the latest NVLink interconnect device. The latter will be available in power servers from International Business Machines Corp. (NYSE: IBM) in 2017.
The New Technology
There are minor, but significant, differences between the two GPUs. The ones for the NVLink will feature 5.3 teraflops, while the Tesla P100 for the PCIe slots will feature 4.7 teraflops. A teraflop is a measure to clock computer graphics processing speed, made up the words tera (meaning trillion) and flop (an acronym for “floating-point operations per second). In simple terms, if the NVLink features 5.3 teraflops, it can process 5.3 trillion calculations in one second. Single-precision performance from the PCIe version should be 9.3 teraflops, while that of the NVLink version is 10.6 teraflops. Some of the other features are similar despite the differences in performance. Each of the two GPUs has an internal memory of 16 GB and a memory bandwidth of 720 GB per second.
The GPUs are created using the latest Pascal architecture which allows them to harness more breakthroughs in performance compared to their predecessors. The Tesla P100 will be incorporated into Nvidia’s proprietary technology known as the CUDA parallel programming framework. The introduction of the company’s mezzanine connector also highlighted the dawn of new card designs from Nvidia, thus deviating from the traditional card structures.
DISCLOSURE: The views and opinions expressed in this article are those of the authors, and do not represent the views of FinancialPress.com. Readers should not consider statements made by the author as formal recommendations and should consult their financial advisor before making any investment decisions. To read our full disclosure, please go to: http://financialpress.com/legal-disclaimer/.