21st Century Super Computing
Supercomputers are ready for your desktop! The technology is here now at a price that is very affordable.
GPU processing fulfills the modern day promise of affordable super computing resources on your desktop. Currently, over a TeraFlop of compute power can be added to a standard desktop personal computer for less than $500.
Many new “High Performance Computing Centers” tout 20 year old cluster technology complete with bandwidth-limited inter-processor links and massive systems management headaches as a supercomputing solution. There is nothing super about these so-called super computers other than their massive size and power consumption. A current 200 quad core processor cluster can manage around 5 Terra Flops (25 GFLOPS per processor) of computing power. This is equivalent to 6 GPUs that would fit easily in two standard desktop cases or in 2U’s of rack space.
While no one was looking,
graphics engine designers developed true supercomputers characterized by
blazingly fast compute engines with large quantities of memory and huge I/O
GPU Computing can enhance even mundane desktop operations with ease. The charts on this page show some initial performance enhancements to Matlab that can be achieved by using a GPU co-processor. In both Matrix Multiplication and Fast Fourier Transforms, the performance increase offered by the GPU is between 5 and 10x. While the Matlab applications do not harness the full power of the GPU, they do give some insight into the power that is available with this approach.
Good things come in small packages.