Fast Games … From 36 HOURS to 15 MINUTES
Xcelerit have got a nice piece of press again in HPCwire. And the article is about how the Xcelerit SDK was used to speed up simulations for cognitive radios dynamically accessing spectrum. This is great.
We were absolutely bowled over when we used the Xcelerit SDK. There are many simulations we need to do at work that really become unwieldy exceptionally quickly. This is especially the case when you want to explore ideas on a network-wide level – in other words when you want to deal with networks that are not made up of three token nodes!
The Xcelerit SDK allows us to convert sequential code to parallel code without needing to have ANY knowledge at all about parallel programming. We just keep focusing on our algorithms. Xcelerit does the rest. And the same code works on multi-core CPUs as well as GPUs. And as you get more hardware, you can simply add more GPUs to your system and get the benefits of the extra hardware and again all without changing the code. In our case the simulation time reduced from 35 hours to 15 minutes using two Nvidia® Tesla™ M2050 GPUs. Our world has changed. We can really work through a lot of ideas in Machine Learning in cognitive radio.
The Xcelerit website talks about ‘supercomputing at your fingertips’. I really understand what that means now that we have worked with it. You get the performance and you get it easily – no specialised knowledge, no need for different code for different hardware, no need for changes when new hardware arrives … really at our fingertips.