Follow this blog:
RSS

Today’s supercomputers: far greater than the sum of their parts

By | November 24, 2009, 2:00 AM PST

Back in the mid-1990s, I had the opportunity to meet Dr. Irving Wladawsky-Berger, then vice president at IBM, who was charged with building a new generation of supercomputers. But Dr. Wladawsky-Berger didn’t go out and spend a huge budget on specialized hardware. Rather, the genius of his approach was to take sets of already-manufactured processors — at the time, RISC processors from IBM’s Unix family — and lashed them together with a high-speed switch to create supercomputing power. A few years later, the follow-on generation of supercomputers was built in a similar fashion from x86 processors running Linux.

Welcome to innovation, 21st Century style. A couple of months ago, I posted some observations about innovation coming from low-cost and conveniently available parts — part of the formula for success for our era: “Good enough” may be good enough when it comes to developing new products and services.

One powerful byproduct that has emerged from this paradigm is supercomputers. The latest and fastest behemoth, as a matter of fact, was built from PC chips from Advanced Micro Devices. The supercomputer, as described by The New York Times’ Ashlee Vance, runs at the Oak Ridge National Laboratory in Tennessee “and can throw hundreds of thousands of processing engines at computing problems.”

“The supercomputing world was long dominated by systems that required specialized chips, memory systems and networking technology. But about 10 years ago, researchers realized they could link thousands of cheaper machines running on mainstream chips and achieve pretty solid performance.”

The idea that many small commodity processors could work together to provide supercomputing power is being seen in other forms, such as grids. For example, the SETI@home project employs individual users across the globe, who donate unused cycles to help sift through data in the search for extraterrestrial intelligence.

The Top500 list of the world’s current top 500 supercomputers was just released, and just about all are built with commonly available commodity processors. In fact, 79% of the world’s fastest supercomputers run on clusters of Intel EM64T processors.

Across all product categories, the high-end brands, typically offered as part of well-crafted and expensive interdependent architectures, will be displaced by more modular approaches offered by commoditizers. In the process, new markets will open as more companies or people have access to more affordable approaches. Or even supercomputers.

Start your week smarter with our weekly e-mail newsletter. It's your cheat sheet for good ideas. Get it.

Joe McKendrick

About Joe McKendrick

Joe McKendrick is a contributing editor for SmartPlanet.

Joe McKendrick

Joe McKendrick

Contributing Editor, Business

Joe McKendrick is an independent analyst who tracks the impact of information technology on management and markets. He is the author of the SOA Manifesto and has written for Forbes, ZDNet and Database Trends & Applications. He holds a degree from Temple University. He is based in Pennsylvania.

Follow him on Twitter.

Joe McKendrick

Joe McKendrick

Joe McKendrick is an independent consultant and editor. Joe has performed project work for the following companies in the IT marketspace: IBM, Systinet/HP, Teradata. He has performed project work for the following organizations in partnership with Unisphere Research (Unisphere Media): IBM, Oracle Corp., International Oracle Users Group, Oracle Applications Users Group, Professional Association for SQL Server, International DB2 Users Group, International Sybase Users Group.

He writes for SmartPlanet and is not an employee of CBS.

If you liked this, don't miss...
2
Comments

Join the conversation!

Follow via:
RSS
0 Votes
+ -
The key to the next generation of super computers
Is a super large, super fast layer 2 switch with physical route addressing built into the header of each packet (no lookup tables or route exploration!) Given enough ports, a sufficient quantity of multi-core processors can be interconnected to provide any conceivable topography and processing concentrations required.
Posted by LarryPTL
24th Nov 2009
0 Votes
+ -
RE: Today's supercomputers: far greater than the sum of their parts
I believe this article (and most modern thinking on "supercomputers") is missing the most basic point in the definition of what a "supercomputer" really is. The article references what really is a super-parallel NOT a super-computer system. A true super computer is one where a single CPU in one step efter another until the end of the program fetches, decodes & executes faster than any other available single CPU system. The true test of a real super computer is not complex multi-input parallel processing weather pattern or atomic level phenomena analysis but rather how fast a step by step problem where the output from step 1 processing is required as input to step 2 etc. can be solved. In simpler terms a super computer system is one with the fastest straight line instruction rate for one CPU, not the greatest aggregate rate for the largest collection of parallel CPUs.
Posted by jcoffey73
24th Nov 2009
Join the conversation
Formatting +
BB Codes - Note: HTML is not supported in forums
  • [b] Bold [/b]
  • [i] Italic [/i]
  • [u] Underline [/u]
  • [s] Strikethrough [/s]
  • [q] "Quote" [/q]
  • [ol][*] 1. Ordered List [/ol]
  • [ul][*] · Unordered List [/ul]
  • [pre] Preformat [/pre]
  • [quote] "Blockquote" [/quote]

Join the SmartPlanet community and join the conversation! Signing up is fast and free. Don't wait -- we want to hear your opinion!