Big Mac at Virginia Tech #1 in USA
2nd most powerful computer on Earth is a Mac
The G5 cluster of the University of Virginia, in the USA, is up and running, under Mac OS X, a version of Panther. Its performances were measured for the first time under the Linpack Benchmark, and scored an amazing 17.6 TFlops !
This result makes this supercomputer the second most powerful one, behind the famed Japanese Earth Simulator, and it is the most powerful of all in the United States. Not bad, when you know it cost a mere 5.2 million dollars, half the estimated price from the competitors during the bid, and most of all, far less than the $100 million that the other current supercomputers did cost.
"Big Mac" (as nicknamed by Terascale staff) will officially join the Top5 Supercomputers during the SC2003 conference in Phoenix, AZ, USA, from the 15th till the 21st November.
P.S. : To help the comparison,
Earth Simulator cost $350 million,
Virginia Tech Mac Cluster Supercomputer $ 5.2 million (1,100 G5 dual 2GHz cluster)
ASCI Q, the (2nd best ...until now), cost $200 million, and
ASCI White (4th position) "only" $100 million.
So, pray tell, are Macs still that expensive?
Mac Supercomputer costs $5.2 million
October 9th 2003
Supercomputers - everyone would love one but with a price tag of around $100 million each they're not easy to come by. But in the United States staff and students at Virginia Tech have built one of the world's most powerful Supercomputers for a cool 5 million dollars. It's all to do with plugging together smaller Mac G5 machines.
Ian Hardy went to Virginia to see the supermachine for himself.
This is the project that has caused heads to turn in the world of Supercomputing. It is, in fact, 1,100 brand new Apple G5 towers placed side by side making it the world's most powerful homebuilt system, capable of 17.6 trillion floating point operations per second, with a combined storage capacity of 176 terabytes.
Pat Arvin, Project Coordinator, Virginia Tech: "Each individual G5 is a dual processor, 2GHZ machine with 4GB of memory. So it's extremely fast." This network is linked using 2900 cables and runs at about 100 times faster than an average corporate network that uses today's top of the line Ethernet connections. But the hard part of making a Supercomputer is stability.
The project's chief architect, Srinidhi Varadarajan, had to write a special program called Déjà Vu to ensure that if any individual computer crashes in the middle of a calculation lasting weeks, if not months, another computer will take over seamlessly. Varadarajan not only helped build the supercomputer, he developed the software package employed in the cluster, Déjà vu, helping bring stability to the computers. Varadarajan said most quality computers fail about once per year and clustering 1100 computers together creates a multiplied failure rate that would be unacceptable. The software transparently transfers a failed application to a new place without the knowledge of the computer, preventing the application from being altered in any way.
With the addition of the supercomputer, the Tech community is open to receiving funds from a proposed National Science Foundation program called The Cyberinfrastructure Funds. The program is scheduled to have a $1 billion annual budget and will offer $100 million to centers like Tech, which house large supercomputer centers supported by the National Science Foundation.
Srinidhi Varadarajan: "This is pretty much like open heart surgery because you're working on a computer and moving an application while it still continues to run. You cannot stop the program, actually, and that's the speciality of this system."
The Supercomputer, unofficially nicknamed Big Mac, was built in just three months. Right from the start there were major hurdles that could only be overcome with significant construction in and around the building. Running 1100 computers in a 3000 square foot area sends the air temperature well over 100 degrees Fahrenheit. In fact the heat is so intense that ordinary air conditioning units would have resulted in 60 mph winds. So specialized heat exchange cooling units were built that pipe chilled water into the facility.
Kevin Shinpaugh, Director of Research/Cluster Computing: "There are two chillers for this project. They're rated 125 tonnes each in cooling capacity, and they pump 750 gallons per minute each. The water is at about 45 degrees Fahrenheit."
The power supply was another huge challenge. The Supercomputer uses the same amount of electricity as 3000 average sized homes. Meanwhile, in early September, the G5s started arriving by the lorry load.
Jason Lockhart, Director of High Performance computing: "The machines were taken off the trucks on their pallets. The pallets were broken down and we processed machines. We did 238 machines in little under two hours, so we were humming along as an assembly line."
A speedy installation was essential because Virginia Tech had to comply with an October deadline set by the National Science Foundation. Missing that deadline would have meant automatic disqualification from the NSF's global Supercomputer rankings, thereby denying the college any chance of competing for top scientific research projects worth hundreds of millions of dollars per year.
Srinidhi Varadaraja: "A system of this size generally sees its most application in what is known as big science research; massive simulations, models, computational engineering systems. Examples of these include things like nanoscale electronics, if you're trying to invent computer chips 30 years from now you're looking at atomic levels with a single atom acting as a switch."
Arguably Virginia Tech has revolutionized the world of Supercomputing with a simplistic setup that can be duplicated around the globe by other institutions.
Hassan Aref, Dean of Engineering: "We're going to document how we did this from start to finish and if you want to build one we'll send them a kit and tell them how to do it."
Supercomputers, like PCs, become old in technology terms within about 3 to 5 years. But whereas you or I would go to a shop and buy one, Virginia tech has to start again from scratch. That's why they are building a new multi million dollar facility opening in 2006 for the next generation of Supercomputer.
1st most powerful computer on Earth
Costs $350-million, achieves 40 Tflops
Distributed Memory Parallel Computing System which 640 processor nodes
interconnected by Single-Stage Crossbar Network
Processor Node: 8 vector processors with shared memory
Peak Performance: 40 Tflops
Total Main Memory: 10 TB
Inter-node data transfer rate: 12.3GB/s x2
Target for Sustained Performance: 5 Tflops using AGCM.
*The goal above was its thousand times' performance of the AGCM in those times (1997).