Performance

Measured Linpack performance

The Linpack Benchmark is used to solve a dense system of linear equations. The Top500 which tracks the 500 most powerful commercially available computer systems known on a semi-annual basis uses a version of the benchmark that allows the user to scale the size of the problem and to optimize the software in order to achieve the best performance for a given machine.

Benchmarks performed by ClearSpeed Technology in August 2006 on a single server with two 3.0 GHz Intel® Xeon® 5160 (Woodcrest) dual core processors system delivered 34 GFLOPS without acceleration. A cluster of four such nodes delivered an impressive136 GFLOPS from its 8 Intel® Xeon® 5160 (Woodcrest) dual core processors while consuming 1,940 Watts of power.

With two Advance accelerator boards in each server, a single node delivered 90 GFLOPS and the cluster performance was increased to over 364 GFLOPS while adding only 200 Watts to the overall power levels, representing over 1GFLOP Linpack per Watt of additional performance. The ClearSpeed accelerated cluster completed the Linpack benchmark run in just 18.4 minutes while using only 40% of the energy required by the non-accelerated cluster which took 48.4 minutes to finish.

To put these results in context, the performance delivered by the four node ClearSpeed accelerated cluster, (a total of 16 CPU cores,) is equivalent to the world's most powerful supercomputer according to the Top500 results from November 1996. That supercomputer was a massive 2048 CPU Hitachi system at the Center For Computational Science at the University Of Tsukuba in Japan that delivered 368.2 GFLOPS.

"Consuming no more power than it takes to turn on the lights in a normal living room, we have increased the performance of the cluster more than two and a half times," said John Gustafson, ClearSpeed chief technical officer for HPC. "With the additional Linpack performance exceeding one GFLOP per Watt and almost perfect scaling, we have demonstrated that ClearSpeed accelerator technology can combine unmatched performance with the economic benefits of reduced energy consumption for HPC clusters."