Nvidia Unveils Next Generation Kepler GPU Compute Engine - hookerseepearrough

Nvidia launched its up-to-the-minute draw of Tesla GPU compute engines at the company's Art Technology League in San Jose today. Ane model shipping directly is supported the existing GK104 chip used in the recently released GTX 680. Dubbed the Tesla K10, the instrument panel delivers as a great deal as 4.6 teraflops of undivided precision floating point performance, or s triad multiplication the single preciseness FP of the older, Fermi-based Nikola Tesla. The card can as wel handle an aggregate memory bandwidth of 320GB per instant. This board is targeted towards oil exploration, signal processing and seismic processing applications.

The more interesting announcement is the Tesla K20. Shapely happening a giant chip with 7.1 billion transistors, the K20 isn't slated for release until Q4. Nvidia's CEO, Jen-Hsun Huang noted that the K20 was the largest, most complex semiconductor device check ever reinforced. It will verisimilar use the Lapp 28nm manufacturing sue used for the GTX 680. The K20 is designed for computationally intensive HPC environments, particularly Finite Component Analysis (FEA), finance and physics applications. It offers three times the double-precision floating point performance of previous generation Tesla products. Additionally to the huge transistor count, the K20 bequeath sport a 384-bit memory interface.

New Features

In addition to improved compute carrying out, the K20 will support several key features to keep the chip busy when being fed work out chores. Hyper-Q increases the number of work queues from a single queue in the previous generation Femtometer chip to 32 work queues. This improves GPU utilization, safekeeping more of the compute cores humming when track parallel compute applications.

Slashing Parallelism behaves like a rather parallel branch prognosticator. When fed tasks, the K20 privy save track of dependent tasks and spawn new compute kernels to complete those tasks, rather than having to request more work from the Central processor.

Huang incontestible a simulation of particles colliding, first starting with the antepenultimate propagation Fermi chip. That GPU could handle 20,000 bodies colliding like a sho at high figure rates. Then he went on to demonstrate real-time clay sculpture of the Andromeda and Milky Way System galaxies colliding – not something we need to worry about for the time being, since IT South Korean won't happen for 3.8 billion years. That simulation ran along a Johan Keple-settled Tesla, showing over 208,000 bodies colliding.

The GPU in the K20, code-named GK110, is expected to personify used in the net Titan supercomputer beingness built at the Oak Rooftree National Laboratory and the Blue Waters system at the General Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign.