Apple technologies have achieved a new milestone in clustering at the University of California in LA (UCLA), and has become the most efficient supercomputer in the Top500 list.
Using a cluster of 128 dual-processor 2GHz Xserve G5s, Mac OS X Server 10.3, a Cisco 6500 Gigabit switch and software from Dauger Research, scientists achieved 1.21 trillion floating-point operations per second.
UCLA’s Plasma Physics Group acquired the Xseves to create the new “Dawson” high-performance parallel computing cluster for plasma physics research.
They’ve conducted some tough calculations on that machine – including one which achieved the benchmark of over a TeraFlop, or one trillion floating-point operations per second, the Altivec Fractal. This was the largest calculation yet performed using Dauger’s Pooch pro software, and put the cluster into the Top 500 Supercomputer list.
The software used also included Pooch Pro and MacMPI, the latest incarnation of AppleSeed, a project begun by physics researchers at UCLA in 1998.
Pooch introduces new technologies to access and run clusters behind firewalls and automatically access job output.
More details on this benchmark research are available online.
UPDATE: Macworld reader Norm LeMieux writes: “In perspective, the 2,200-unit cluster at Virginia Tech (System X) is listed at 20.24Tflops. This divides down to 9.2Gflops per unit (dual 2.3GHz Xserves), while the UCLA 128-unit cluster divides its 1.21Tflop performance down to 9.6Gflops per unit (dual 2GHz Xserves).
“This makes it more “efficient” than any other supercomputer in the Top 10 (for which I did this calculation). And we’re not even talking about cost yet.”