Monday, November 12th 2012

NVIDIA Unveils World's Fastest, Most Efficient Accelerators

NVIDIA today unveiled the NVIDIA Tesla K20 family of GPU accelerators, the highest performance, most efficient accelerators ever built, and the technology powering Titan, the world's fastest supercomputer according to the TOP500 list released this morning at the SC12 supercomputing conference.

Armed with 18,688 NVIDIA Tesla K20X GPU accelerators, the Titan supercomputer at Oak Ridge National Laboratory in Oak Ridge, Tenn. seized the No. 1 supercomputer ranking in the world from Lawrence Livermore National Laboratory's Sequoia system with a performance record of 17.59 petaflops as measured by the LINPACK benchmark.

Tesla K20 - Performance, Energy-Efficiency Leadership
Based on the revolutionary NVIDIA Kepler compute architecture, the new Tesla K20 family features the Tesla K20X accelerator, the flagship of NVIDIA's Tesla accelerated computing product line.

Providing the highest computing performance ever available in a single processor, the K20X provides tenfold application acceleration when paired with leading CPUs. It surpasses all other processors on two common measures of computational performance -- 3.95 teraflops single-precision and 1.31 teraflops double-precision peak floating point performance.

The new family also includes the Tesla K20 accelerator, which provides 3.52 teraflops of single-precision and 1.17 teraflops of double-precision peak performance. Tesla K20X and K20 GPU accelerators representing more than 30 petaflops of performance have already been delivered in the last 30 days. This is equivalent to the computational performance of last year's 10 fastest supercomputers combined.

"We are taking advantage of NVIDIA GPU architectures to significantly accelerate simulations in such diverse areas as climate and meteorology, seismology, astrophysics, fluid mechanics, materials science, and molecular biophysics," said Dr. Thomas Schulthess, professor of computational physics at ETH Zurich and director of the Swiss National Supercomputing Center. "The K20 family of accelerators represents a leap forward in computing compared to NVIDIA's prior Fermi architecture, enhancing productivity and enabling us potentially to achieve new insights that previously were impossible."

Additional early customers include: Clemson University, Indiana University, Thomas Jefferson National Accelerator Facility (Jefferson Lab), King Abdullah University of Science and Technology (KAUST), National Center for Supercomputing Applications (NCSA), National Oceanic and Atmospheric Administration (NOAA), Oak Ridge National Laboratory (ORNL), University of Southern California (USC), and Shanghai Jiao Tong University (SJTU).

Energy-Efficiency for "Greener" Data Centers
The Tesla K20X GPU accelerator delivers three times higher energy efficiency than previous-generation GPU accelerators and widens the efficiency advantage compared to CPUs.

Using Tesla K20X accelerators, Oak Ridge's Titan achieved 2,142.77 megaflops of performance per watt, which surpasses the energy efficiency of the No. 1 system on the most recent Green500 list of the world's most energy-efficient supercomputers.

Fastest on Broadest Range of Data Center Applications
The Tesla K20 family accelerates the broadest range of scientific, engineering and commercial high performance computing and data center applications. Today, more than 200 software applications take advantage of GPU-acceleration, representing a 60 percent increase in less than a year.

When Tesla K20X GPU accelerators are added to servers with Intel Sandy Bridge CPUs, many applications are accelerated up to 10x or more, including:
● MATLAB (engineering) - 18.1 times faster
● Chroma (physics) - 17.9 times faster
● SPECFEM3D (earth science) - 10.5 times faster
● AMBER (molecular dynamics) - 8.2 times faster

More information about the Tesla K20 GPU accelerators is available at NVIDIA booth 2217 at SC12, Nov. 12-15, and on the NVIDIA high performance computing website. Users can also try the Tesla K20 accelerator for free on remotely hosted clusters. Visit the GPU Test Drive website for more information.
Add your own comment

11 Comments on NVIDIA Unveils World's Fastest, Most Efficient Accelerators

Some serious hardware we have here.
Posted on Reply
so when we get to see GK110 based geforce?
Posted on Reply
so they had to disable a SMX in teh server version huh, maybe its because they're running it passively.
Posted on Reply
renz496 said:
so when we get to see GK110 based geforce?
IMHO never. I think that it's quite obvious that Kepler gaming parts suck in compute (compared to Fermi) but are good in gaming. I imagine that in GK110 they only went for compute performance.

But like I said, this is the way I see it.
Posted on Reply
For anyone wondering here is the website for the Titan, its a cool computer wish I could use just one rack...

Quote: The Oak Ridge Leadership Computing Facility (OLCF) is home to Titan, the world's most powerful supercomputer for open science with a theoretical peak performance exceeding 20 petaflops (quadrillion calculations per second). That kind of computational capability—almost unimaginable—is on par with each of the world's 7 billion people being able to carry out 3 million calculations per second. Image courtesy Oak Ridge National Laboratory

Posted on Reply
repman244 said:
IMHO never. I think that it's quite obvious that Kepler gaming parts suck in compute (compared to Fermi) but are good in gaming. I imagine that in GK110 they only went for compute performance.
But like I said, this is the way I see it.
There are bound to be GPU's that don't fit within the voltage/wattage (225-235W) spec for Tesla - and likely Quadro. I could quite easily see Nvidia putting a out a limited (availability) GeForce SKU with those parts, if only for PR purposes.

Estimates for gaming performance put the GK110 at around fifty percent greater of the GK104. Assuming that the 15% speedup for GK114 and HD8900 series over the present parts holds true -likely given die/TDP/wafer cost constraints - a GK110 GeForce would make a compelling marketing tool (esp. if the GK114 falls short of a HD 8970's performance).
Posted on Reply
It's funny how this one card holds much more computing power than the fastest supercomputer of the time 17 years ago had. (1310 Gflops of this card vs 614 Gflops peak of the fastest supercomputer at the time)

Makes me realize how much power we have at our hands these days...
Posted on Reply
With all these Press Release about "Titan" you'd think there would be a mention of the people who built the darn thing.

CRAY Inc. :respect:

Just saying...
Posted on Reply
Add your own comment