When the latest list over the world’s 500 fastest supercomputer has now been released it turns out that China, as expected, has the world’s most powerful supercomputer in the form of Tianhe-1A. And that the NVIDIA Tesla cards the system uses has taken a firm hold of the top 5.
Three of the five fastest supercomputers uses NVIDIA’s lstest Fermi Tesla architecture and the Asian companies seem to be the first to understand the advantages of a GPU-based solution.
Tianhe-1A is the system that uses the power to the fullest and with a en maximal performance of 2,566 Teraflops it takes the lead, but also the world’s third and fourth fastest supercomputeruses Tesla graphics cards. Nebulae that is also found in China takes third place with 1,271 Teraflops, while TSUBAME 2.0, located in Japan, follows closely with 1,192 Teraflops.
Tianhe-1A houses 7,168 Tesla M2050 GPUs
Europe’s fastest supercomputer, Tera-100, climbs up to sixth place with a maximal performance of 1,050 Teraflops – but without GPU support.
Noteworthy is that the three partly GPU-based supercomputers has a relatively low power consumption. Tianhe-1A consume at most 4,040 kilowatt, which is still a lot of juice, but with considerably higher performance than Cray Jaguar on second place that consumes 6,960 kilowatt at most, you can’t deny the efficiency of the GPUs.
The entire top500 list can be found here.