College Prof, Student Build Inexpensive, Desktop-Size Supercomputer
The biggest hurdle was designing and building a system for less than $2,500, which was all they had from Calvin College in Grand Rapids, Mich.
A computer science professor with the help of one of his students has built a desktop PC-size supercomputer with off-the-shelf hardware that costs less than $1,300.
Joel Adams, computer science professor at Calvin College in Grand Rapids, Mich., and then-senior Tim Brom set out in January to build a personal supercomputer that would fit next to a desk, or could be taken on a plane as checked baggage. The biggest hurdle was in designing and building a system for less than $2,500, which was all they had from the college.
"There's lots of off-the-shelf hardware to choose from," Adams told InformationWeek. "But you need to balance your processor speeds, the amount of memory you have, and the network bandwidth all within a measly $2,500 budget. That's the challenge."
The researchers got their inspiration from Little Fe, another academic effort to build a personal supercomputer. Little Fe built its first system in 2005, but Adams wanted to significantly increase the price-performance ratio. This was possible because of the better hardware available today.
Rather than use single-core processors and Fast Ethernet connections of 100 Mbps used in the original Little Fe project, Adams chose dual-core Athlon 64 processors from Advanced Micro Devices, and Gigabit Ethernet technology that moves data between hardware at a gigabit a second. With the hardware advancements, and lots of fine-tuning, Adams and Brom built Microwulf, which is capable of processing 26.25 billion double-precision floating-point instructions per second.
Microwulf's processing power is twice as fast as Deep Blue, which was the IBM-created supercomputer that beat world chess champion Gary Kasparov in 1997. The price tag for Deep Blue, then the state of the art, was $5 million.
When completed in March for $2,470, Microwulf had a price-performance ratio of $94.10 per gigaflop. In August, the same parts could be bought for less than $1,256 for a price-performance ratio of $47.84 per gigaflop, Adams said. Those numbers mean Adams and Brom, who is now a graduate student at the University of Kentucky, met their goal.
"We wanted to have big enough performance, so it would be a legitimate research machine," Adams said. "We wanted to provide a personal platform for building parallel applications and testing their scalability before moving them to a big cluster. By doing a lot of the initial tuning on this personal system, it would take less time to tune [applications] on the big cluster."
Large computer clusters that act as one system are used today to drive the most data-intensive applications. Those systems, which are more than 100 times faster than Microwulf, are used by the National Weather Service to process meteorological data and by the U.S. Missile Defense Agency to simulate nuclear tests.
Microwulf and Little Fe are considered Beowulf clusters, which are a group of networked computers that run open-source software and work in parallel. Microwulf relies on four dual-core motherboards connected by an eight-port Gigabit Ethernet switch. The components form a three-tiered system that looks like a triple-decker sandwich. Microwulf is 11 inches by 12 inches, and is 17 inches high.
Blueprints for Microwulf are available on Cluster Monkey, a site dedicated to cluster computing. Adams and Brom don't intend to try to make any money off of their work, and have published the technical details for anyone to use.
Rather than pursue anything commercial, Adams said he's on to his next project. "We've got a grant from the National Science Foundation to build a big cluster, so that's what's on my plate."
5 Top Federal Initiatives For 2015As InformationWeek Government readers were busy firming up their fiscal year 2015 budgets, we asked them to rate more than 30 IT initiatives in terms of importance and current leadership focus. No surprise, among more than 30 options, security is No. 1. After that, things get less predictable.