Cray unveils Cray XC30 supercomputer, capable of scaling to 100 petaflops
November 12, 2012
Cray XC30 supercomputer (credit: Cray Inc.)
Cray Inc. has launched the Cray XC30 supercomputer, previously code-named “Cascade,” designed to scale high performance computing (HPC) workloads of more than 100 petaflops, with more than one million cores.
Cray did not specify whether the 100 petaflops was Rpeak or Rmax, or when a 100 petaflops installation might be planned.
China’s Guangzhou Supercomputing Center also recently announced the development of a supercomputer capable of 100 petaflops peak performance: the Tianhe-2 supercomputer, due to launch in 2015.
Developed in conjunction with the U.S. Defense Advanced Research Projects Agency, the Cray XC30 combines the new Aries interconnect, Intel Xeon processors, Cray’s fully-integrated software environment, and innovative power and cooling technologies.
Die shot of the Aries interconnect chip (credit: Cray Inc.)
Several leading HPC centers have signed contracts to purchase Cray XC30 supercomputers, including:
- The Swiss National Supercomputing Centre (CSCS) in Lugano, Switzerland
- The Pawsey Centre in Perth, Australia, owned by CSIRO and operated by iVEC
- The Finnish IT Center for Science Ltd. (CSC)
- The Department of Energy’s National Energy Research Scientific Computing Center (NERSC) in Berkeley, Calif.
- The Academic Center for Computing and Media Studies (ACCMS) at Kyoto University in Kyoto, Japan
- The University of Stuttgart’s High Performance Computing Center Stuttgart (HLRS) in Germany
The Cray XC30 will utilize the Intel Xeon processors E5-2600 product family. Future versions of the Cray XC family of supercomputers will be available with the new Intel Xeon Phi coprocessors and NVIDIA Tesla GPUs based on the next-generation NVIDIA Kepler GPU computing architecture.
Early shipments of the Cray XC30 are starting now, and systems are expected to be widely available in first quarter of 2013.
“Cray is a leader in the high-end of the supercomputing industry, and the Cray XC30 system promises to continue the Company’s strong standing in the market for designing, building and installing leadership-class supercomputers, such as the ‘Titan’ system at Oak Ridge National Laboratory and the ‘Blue Waters’ supercomputer at the University of Illinois’ National Center for Supercomputing Applications,” said Earl Joseph, IDC program vice president for HPC. “The Cray XC30 supercomputer also advances Cray’s Adaptive Supercomputing vision, which aims to boost application performance for their customers by exploiting hybrid processing.”
The Cray XC30 supercomputer is made possible in part by Cray’s participation in the Defense Advanced Research Projects Agency’s (DARPA) High Productivity Computing Systems program.
Comments (16)
by Sea Bass
Good point. I believe nature accomplishes this feat because of how it processes information. I think transmission between neurons do at least two things at the same time:
Encourage the propagation of the signal by establishing more neuronal connections to pass the information.
Suppress the signal by de-coupling of neuronal connections to prevent the passing of information.
As a Turing machine, the bridge and gaps created between neurons could represent the two states of information (1′s and 0′s).
by Doug
for more detail, go to http://www.top500.org/ Based on 20 years of tracking, this site states “Our projection also shows that the first Exaflop/s computer will enter the TOP500 list in 2019, and only one year later, in 2020, there will be the first notebooks with a performance of 100 Teraflop/s.”
by Erik
We need better hardware.
That they are going to build a 100 petaflops machine on dedicated graphics hardware in a few places in three years doesn’t make it a software thing.
The average computer scientist has about 100 gigaflops on his desktop computer today. That is what matters.
Let’s say you have an algorithm and it takes an hour to execute on 100 petaflop computer. Doing the same number of calculations on that desktop computer will take more than hundred years!
Also, don’t forget it takes a human brain many years to learn how to
talk and reason properly.
by PirateRo
Wow……..
by Gorden Russell
So just how powerful does a computer have to be to become an AI?
by Daworox
Depends on a software ;)
by Mr.X
@Gorden:
It won’t become an AI because of power, but because of better software.It’s the software that’s lacking.
by renzo canepari
one hundred petaflops is 10 to the 17th. The highest number is saw in Singularity was 10 to the 19th. We’re getting real close to the requisite hardware–if Ray Kurzweil is right.
by Gorden Russell
Thanks guys.
by Gabriel
In his “growth of supercomputer power” chart, Kurzweil notably predicts that supercomputers will have the power “required for human brain functional simulation” by 2013…
by melajara
Please, before attempting this, let’s adequately simulate a bee brain with its sensorium. I’m not sure any team is able to do it yet, notwithstanding the forthcoming availability of 100 petaflops computers!
by Dr.Pratt
Your brain allows consciousness, it does not creat it. There is almost no power at all, in the human brain.
by Mr.X
@Dr.Pratt: You’re not a doctor, I hope.
by Gabor
This question is very tricky because we do need more power to get strong AI and we do need (much) better software, too, but I believe most people ignore another two ingredients to achieve a high level consciousness. One is parallelism and the other is all the above achieved in a relatively small space. The first one is easy as all we need is more cores working in unison, maybe somewhere on the scale of our own number of neurons (100 Billion or more). Simulating 100 Billion neurons on a computer that works in series for the most part will create an extremely smart “Watson” but not an intelligent consciousness because the delay between the processes do to the speed limit (light speed). And the light speed is exactly why we need to achieve this parallelism within a relatively small space that is my second point. Then again, your question was AI and not consciousness. On AI, I suppose you mean strong AI which I believe we are very close to have enough “power” just mostly lacking in software.
by melajara
What we are completely lacking is a reflective model of computation ensuring deep learning (and for robots, sensorimotor scheme compositions) through largely self programmed subservient modules.
This is yet another shortcoming of current IA, ignoring the grounding problem and the formidable pruning it allows in combinatorial composition of internal representations.
So much brute power, so shallow thought invested.
Who are the Turing or von Neumann of our generation?
by Bri
Again I’ve got to bring up my jumping spider analogy. Such a small brain and yet it processes so much information. They see amazingly well. They hunt their food, so they identify things very well. They use eight legs to get around so they have lots of movement processing. All in a brain case less than the size of a pin head. I don’t think we need greater power. I think we need a radical change in architecture and programing. I also think that the fundamentals have to be somehow simple. I think the whole system of life is based more on vibrational frequencies. Maybe something akin to a holographic computer. The different brainwaves acting like carrier frequencies in FM circuits. Neurons that wire together, fire together. That’s radically different than how our computers work today.