Monday, September 25, 2017

CST STUDIO SUITE and GPU computing

The high memory bandwidth and parallel processing abilities of GPU cards mean that
GPU computing can provide significant simulation speed advantages over conventional
CPU computing. CST STUDIO SUITE® supports GPU computing in the transient solver,
integral equation solver, asymptotic solver, particle-in-cell (PIC) solver, and in the new conjugated heat transfer solver using NVIDIA Tesla® or high-end Quadro GPU cards.

All GPU-capable solvers now support the newest generation Pascal series Tesla cards which are dedicated for server-class systems; selected Quadro cards are supported for workstations. 

The GPU cards of the Pascal generation (Tesla P100 and Quadro GP100) offer an impressive speed-up over the high-end models of the previous Kepler series generation as the following diagram shows

What makes the performance improvement as compared to CPU even more impressive is the fact that the speedup numbers in the diagram are achieved when comparing to a really high end CPU configuration with dual Xeon Gold 6148 CPUs. Thus, the new Pascal generation is nowadays the recommended platform for GPU computing with CST STUDIO SUITE.      


Please see our list of supported GPU cards for a complete overview of GPU devices supported by CST STUDIO SUITE.
Due to the very good scalability of our GPU implementations with the number of GPU devices, simulations can be accelerated even further by adding more GPU cards to a host system as the following diagram shows for a transient simulation with 82 million mesh cells. The reference is again provided by a system with dual Xeon Gold 6148 CPUs.





Dr.-Ing. Felix WOLFHEIMER
Manager for High Performance Computing 
CST-Computer Simulation Technology












No comments:

Post a Comment