More CAE applications are supporting GPU computing. Siemens and Dassault Systemes use OpenCL to optimize simulation problems and deliver higher performance.
CAE applications are critical to design and engineering. And simulations demand computing resources. When a workstation has a powerful GPU as well as a powerful CPU, optimized simulations can deliver results faster.
Professional GPU vendors have partnered for decades with software vendors. There is a beneficial synergy when application developers have access to graphics experts. Most recently, these partnerships have moved beyond graphics and into simulation and other computing areas.
OpenCL is an open, industry standard for programming non-graphics computing on GPUs. Companies like AMD support this standard fully with their professional FirePro graphics line. Nastran and Abaqus are among the simulation applications which leverage OpenCL and FirePro GPUs to accelerate simulation calculations.
GPU computing can dramatically improve simulation performance, but it depends on the problem to solve. The problem must be highly parallel, and at this time, non-recursive. Solving large systems of non-linear equations are typical problems which benefit from GPU acceleration.
For application developers, fast floating point performance is not sufficient. First of all, good programming tools are needed to support development of code running on GPUs. OpenCL is an industry standard which makes this possible. Functionally, a fast GPU for calculations is also not sufficient. First, simulations almost always require double precision floating point calculations. Single precision performance is nearly irrelevant.
Second, simulations need large graphics memory. This allows for better optimizations as larger data sets can be calculated locally on the graphics card which reduces communication overhead between the GPU and the system.
Third, and related to the previous point, memory bandwidth performance between the system and the GPU is critical.
In the case of Simulia Abaqus and Siemens NX Nastran, AMD offers the FirePro W8100 and FirePro W9100. Both have fast GPUs, support for PCIe Gen 3, and fast memory bandwidth performance. The FirePro W9100 reaches 2.62 TFLOPS of double precision computing performance which is untouched by any other GPU on the market today. It's 16 GB of GDDR5 memory is double that available on the FirePro W8100 and more than any GPU from NVIDIA.
This brings together three important aspects for optimizing engineering workflows:
partnerships between graphics vendors and software vendors
developer tools like OpenCL
fast GPUs with the right features to support simulation applications
In summary, we see more applications taking advantage of the full power of a workstation. Technology advances with generational jumps, but technology is also transformed as partnerships bring new possibilities to market.