The paradigm of supercomputing has shifted rapidly during the past decade. Ten years ago when we heard “NVIDIA”, immediately we associated the brand name with computer graphics, games and animation. NVIDIA’s breakthroughs in graphics processing unit (GPU) technology make supercomputing inexpensive and widely accessible nowadays. In addition to its visual computing leadership, NVIDIA also strives for green computing where its hardware design aims at the best performance per watt. More than eight teraflops of computing power can be achieved on an NVIDIA Tesla K80 that consumes less than 300 watts of electricity.
To harness the power of GPUs for electromagnetics simulation, ANSYS released the GPU-accelerated HFSS Transient in 2014. The new solver is tailored for customers who want to solve electrically large structures and high-order finite-element meshes more efficiently. Instead of upgrading to workstations with high-end CPUs, customers can add high-end GPU cards to their existing machines and enjoy a performance boost right away. The GPU-accelerated solver can typically achieve two-times speedup on one NVIDIA Tesla K20 versus eight cores of Intel Xeon X5675. In benchmarks of 15 examples, a maximum speedup of 5.2 times was achieved.
To further illustrate the application of GPU acceleration, this image shows the transient field analysis of a smart phone, which is tessellated into 1,093,376 tetrahedrons by hp-adaptive mesh refinement. The speedup of one NVIDIA Tesla K40 versus eight cores of Intel Xeon E5-2687W is 4.8 times, and the GPU RAM requirement is 5.99 GB.
You can find more examples in the ANSYS Advantage article entitled GPUs Speedup the Solution of Complex Electromagnetic Simulation
To learn more about the capability and technology of the GPU-accelerated HFSS Transient, I invite you to attend my GTC presentation in San Jose, California on March 20th. For those who cannot attend the conference, the recorded presentation will be posted on the GTC website afterwards and I will be able to answer your questions anytime.