IBM to build fastest computer on the planet (again)

February 9, 2009

by Mark Ollig

How much “faster” can combined petascale supercomputer processors be made to process information?

Last June I wrote a column about the IBM “Roadrunner” supercomputer, which was operating at 1 petaflop (one thousand trillion floating point operations per second).

In November of 2008, the Roadrunner was tested again and obtained a peak benchmark performance of 1.105 petaflops.

The Roadrunner supercomputer is still the fastest computer ever built.

I thought 1 petaflop was fast enough, but the US government has again selected IBM to build an even faster computer.

How about a 20 petaflops supercomputer?

According to IBM, the new “Sequoia” supercomputer will be able to achieve mind-boggling performance speeds of up to 20 petaflops, or 20,000 trillion calculations a second.

It will use a total of 1.6 million microprocessor cores aided by 1.6 petabytes worth of memory storage.

Let’s break down these “bits & bytes,” which here represent disk storage, as stated in the IBM dictionary:

• 1 Peta (quadrillion) byte is approximately 1,000 Tera (trillion) bytes.

• 1 Terabyte is approximately 1000 Giga (billion) bytes.

• 1,000 Mega (million) bytes is approximately 1 Gigabyte.

• 1,000 Kilo (thousand) bytes is approximately 1 Megabyte.

• 8 bits of information is equal to 1 byte.

One bit of disk storage is the smallest unit of data that a computer uses. It is used to represent two states of information – think of it as yes or no, or a state of being off or on.

One byte can represent up to 256 different states of information, like numbers or a combination of numbers and letters.

One byte of storage is required for one single character.

Ten bytes could be equal to a word, and 100 bytes would equal an average sentence.

If you are interested in understanding more of these computing storage terms, including how large a “yottabyte” file is, visit this excellent reference I found at http://www.whatsabyte.com.

While writing this column I was also watching “C-Span.” One of the speakers, Sen. John Thune of South Dakota, was describing how $1 trillion would be equal to placing single $100 bills on top of each other 689 miles high.

When your humble columnist heard the word “trillion,” I thought it apropos to mention it here since we are talking about large numbers.

The senator even had a “cartoonist” like chart displaying a colorful drawing of the $100 bills neatly stacked in a single row towering high above the Earth and into the darkness of space.

Given my minimal understanding of Planetology and Newton’s simplified gravitational theories, I do not believe the $100 bills would stay this neatly stacked – particularly with the effects of the Earth’s rotation, trade winds, orbiting satellites, and things like that.

Sen. Thune’s presentation reminded me of those popular charts Ross Perot used when he ran for president some years back.

In the end, it appeared Sen. Thune believed he made his point.

Getting back to the Sequoia, IBM estimates the computing power of this new supercomputer will be greater than all of the current computers on the “Top 500” supercomputer rankings list combined.

IBM said the Sequoia will be used for research on astronomy, energy, human genome or DNA research, and global climate change. In addition, this new supercomputer will also provide much improved local weather forecasting accuracy.

I learned the main purpose of this new supercomputer will be for the US government to run complex computer simulations for nuclear weapons research.

Here are some other interesting numbers to ponder.

It would take 120 billion people – almost 20 times as many people currently on Earth – constantly using calculators for 50 years to process what the Sequoia could accomplish in a single day.

The everyday calculator we use operates at about 10 floating point operations per second (flop/s). Each time we perform a calculation, it only requires a single operation. Any response time below 0.1 seconds from the calculator seems instantaneous to us.

The cost to build the Sequoia has not been made public, but I have read where it is likely to be well over $100 million.

The Sequoia’s completion date is set for 2012.

The new supercomputer will be housed at the US Department of Energy’s Lawrence Livermore National Laboratory in California.

This story also has a local connection – the Sequoia supercomputer will be built by IBM in their Rochester Min. “BlueGene” facility.

This week’s “Web Site of The Week” forum at the Herald Journal web site will have video, pictures, and information about the Sequoia and other supercomputers.