Supercomputers are still in the entry stage of using graphics circuits for speeding up calculations. Australian Commonwealth Scientific and Research Organization (CSIRO) has now presented a new supercomputer that will put emphasis on using graphics circuits from NVIDIA. Except from more than one hundred Intel Xeon processors the supercomputer will also house 50 Tesla GPUs connected to a 80 Terabyte storage unit from Hitachi.
The decision to use NVIDIA’s Tesla GPUs will according to CSIRO result in 10 to 100 times better performance when crunching scientific data.
Despite relatively few circuits the supercomputer will achieve more than 200 TFLOPS, which is not to frown upon. By using graphics circuits the graphics circuits the computer can match the performance twice as big, which is a clear indicator of the efficiency of the system.
NVIDIA hopes that its Fermi architecture will result in a nice boost in the right direction and it looks like more companies are starting to get ready to adopt the Tesla family for supercomputing.