Page 43

EN-July2017-eMag6

ing on graphics processing units.  Until fairly recently GPUs were not generally accessible by any software other than graphics software. With the arrival of the NVIDIA Tesla GPU all that changed. Tesla GPUs are based on CUDA, NVIDIA’s computing architecture that enables its GPUs to be programmed using industry-standard programming languages like C++ and C#, opening up massively parallel processing power to a broad range of computing applications beyond traditional graphics. With power-efficient cores and increasingly fast access to memory, GPUs are well suited to accelerate many CAE simulations. Benchmarking of the new ANSYS Mechanical implementation has shown that double precision computations of a typical workload, can be performed on a GPU up to twice as fast as a traditional CPU. This technology milestone was demonstrated on the NVIDIA Tesla platform with a variety of customer-relevant models. “This initial development for GPU computing demonstrates our focus on evolving ANSYS software to take advantage of important technology trends in high-performance computing. HPC is a rapidly changing technology arena and also a key enabler of Simulation Driven Product Development,” said Dipankar Choudhury, vice president of corporate product strategy and planning at ANSYS. “We work to achieve optimised software performance, across the full spectrum of HPC technologies, so that our customers get maximum value from their investment in HPC. Here, our technical collaboration with NVIDIA has resulted in a significant benefit for our mutual customers.” MSC.Software’s Marc finite element analysis software, has demonstrated performance gains, with Tesla GPUs, of up to 5x for typical complex models compared with the latest dual-core CPUs. “Our customers will achieve new levels of simulation fidelity and product innovation using this breakthrough technology,” says Ash Munshi, CEO of MSC.Software. “This successful collaboration with NVIDIA is the beginning of an alliance that will ultimately add value across a broad range of simulation products from MSC.Software.” The GPU solver work at MSC.Software is being carried out on the advanced non-linear solver MSC.Marc and it will be available soon. “NVIDIA Tesla GPUs have quickly gained a reputation in the HPC industry for delivering dramatic changes to workflow, resulting in significantly reduced computational time across complex problems,” said Andrew Cresci, general manager of vertical solutions at NVIDIA. “The combined processing power of Tesla GPUs and ANSYS tools accelerates time to insight and helps ANSYS customers deliver innovative products that can consistently exceed market expectations.” HOUSEKEEING HEADACHE The use of GPUs for CAE simulation also causes the software vendors a housekeeping headache. Generally speaking the software licensing is tied to the number of processors in use in the traditional CPU sense ie. dual-core or quad-core. So a new and different pricing model will need to be developed for GPUs that can house hundreds and sometimes thousands of simple processors. The GPU accelerator capability accelerates only the shared-memory equation solvers that support the usage of a GPU. In the case of Ansys these are the sparse direct and preconditioned conjugate gradient (PCG)/Jacobi conjugate gradient (JCG) iterative solvers. This includes the use of block Lanczos and PCG Lanczos solvers in an eigenvalue buckling or mode frequency analysis. Other equation solvers will continue to run on only the CPU and will not see any performance benefit when using the GPU accelerator capability. As yet much of the simulation using GPUs is limited to linear and static stress, but there is a lot of work being done especially is specialist areas such as Formula One CFD. Such is the competitive advantage gained by this technology that the practitioners and protagonists are still very reluctant to talk about the details. As more work is done, more complex analyses will eventually be accommodated. www.engineeringnews.co.nz 43


EN-July2017-eMag6
To see the actual publication please follow the link above