NVIDIA Pioneers New Conventional for Great Performance Handling With tesla GPUs Designed on Kepler Architecture
New Structure Triples Energy Performance, Creates GPU-Accelerated Handling Techniques Simpler to Program
GPU Technological innovation Meeting - NVIDIA these days revealed a new family of Tesla® GPUs depending on the brand new NVIDIA® Kepler™ GPU computing architecture, which makes GPU-accelerated computing simpler and more available for a greater variety of high performance computing (HPC) medical and specialized programs.
The new NVIDIA tesla K10 and K20 GPUs are computing accelerators created to deal with the most complicated HPC issues in the world. Designed with an excessive concentrate on high performance and excessive efficiency, Kepler is three periods as effective as its forerunner, the NVIDIA Fermi™ architecture, which itself founded a new standard for similar computing when presented two years ago.
"Fermi was a significant advancement in computing," said Expenses Dally, primary researcher and mature v. p. of analysis at NVIDIA. "It founded GPU-accelerated computing in the top level of high performance computing and drawn tons of designers to the GPU computing foundation. Kepler will be similarly troublesome, developing GPUs generally into specialized computing, due to their convenience of use, wide usefulness and efficiency."
The tesla K10 and K20 GPUs were presented at the GPU Technological innovation Meeting (GTC), as part of a sequence of reports from NVIDIA, all of which can be utilized in the GTC online media room.
NVIDIA created set of modern structural technological innovation that make the Kepler GPUs high doing and extremely cost effective, as well as more appropriate to a greater set of designers and programs. Among the significant enhancements are:
SMX Loading Multiprocessor -- The primary source of every GPU, the SMX streaming multiprocessor was remodeled from the earth up for top performance as well as efficiency. It provides up to three periods more performance per w than the Fermi streaming multiprocessor, making it possible to develop a supercomputer that provides one petaflop of computing performance in just 10 host shelves. SMX's energy efficiency was obtained by improving its number of CUDA® architecture cores by four periods, while decreasing time rate of each primary, power-gating areas of the GPU when lazy and improving the GPU area dedicated to parallel-processing cores instead of management sense.
Dynamic Parallelism -- This ability allows GPU strings to dynamically create new strings, enabling the GPU to evolve dynamically to the information. It significantly makes easier similar coding, empowering GPU rate of a greater set of well-known methods, such as flexible capable processing, quick multipole techniques and multigrid techniques.
Hyper-Q -- This allows several CPU cores to at the same time use the CUDA architecture cores on 1 Kepler GPU. This considerably improves GPU usage, cutting CPU lazy periods and improving programmability. Hyper-Q is perfect for group programs that use MPI.
"We designed Kepler with an eye towards three things: performance, efficiency and availability," said Jonah Alben, mature v. p. of GPU Technological innovation and key designer of Kepler at NVIDIA. "It symbolizes an important landmark in GPU-accelerated computing and should nurture the next trend of improvements in computational analysis."
NVIDIA Nikolatesla K10 and K20 GPUs
The NVIDIA Nikola tesla K10 GPU provides the best throughput for indication, picture and seismic processing programs. Enhanced for clients in oil and gas discovery and the immunity market, 1 tesla K10 reduce panel functions two GK104 Kepler GPUs that provide an total performance of 4.58 teraflops of optimum single-precision sailing point and 320 GB per second storage information.
The tesla K20 is depending on the GK110 Kepler GPU. This GPU provides three times more twice perfection when in comparison to Fermi architecture-based tesla products and it can handle the Hyper-Q and powerful parallelism abilities. The GK110 GPU is supposed to be included into the new Powerhouse supercomputer at the Oak Variety Nationwide Clinical in Tn and the Red Ocean system at the Nationwide Heart for Supercomputing Programs at the School of Il at Urbana-Champaign.
"In the two decades since Fermi was released, multiple processing has become a commonly implemented way to accomplish greater efficiency for a number of crucial HPC applications," said Earl C. John, system v. p. of High-Performance Computing at IDC. "Over the next two decades, we anticipate that GPUs will be significantly used to provide greater efficiency on many applications.