Thursday, November 14, 2013

Accelerate simulations with the Nvidia Quadro K6000

Nvidia® recently released a new high-end graphics card in the Kepler series, the Nvidia Quadro® K6000. This card has a large amount of onboard RAM (12 GB), making it very suitable for large and demanding tasks. Starting with CST STUDIO SUITE 2013 Service Pack 4, all solvers in CST STUDIO SUITE with GPU computing acceleration now support the Quadro K6000.

To test its performance, we carried out a series of simulation benchmarks with a Quadro K6000, comparing it to both to another Nvidia GPU card, the Nvidia Tesla® K20c, and to an equivalent system with two Xeon E5-2650 CPUs.

As shown in the graph below, both GPU cards offered a significant benefit when compared to CPU simulation. However, for larger models, with tens of millions of mesh cells, the Quadro K6000 showed a performance improvement of between 30% and 35% compared to the Tesla K20c, giving a four-fold speedup compared to the CPU simulation. The benchmarks were performed on a ProViz W60 workstation with the time domain solver in CST STUDIO SUITE 2013. The hardware for this benchmark was kindly provided by CADnetwork.


The speedup compared to CPU computing for the Tesla K20c and the Quadro K6000 when simulating a multi-port IC package (pictured).


No comments:

Post a Comment