ANSYS CFD is on the verge of a second renaissance in high-performance computing (HPC). The first, spanning more than a decade, has seen tremendous leaps in both the depth and breadth of HPC capabilities. Depth (or heights, rather) in the size of the scalable clusters — first 1000s, then 10K, and recently 100K core counts — and breadth of coverage across solvers, physics, post-processing, even file I/O, covered the gamut of high-performance simulations. The trend, in fact, is exponential, as evident in this chart, and spans many years of ANSYS Fluent software releases. While there are other impressive scientific scalability demonstrations, ANSYS Fluent set the standard for industrial HPC CFD simulations.
Meanwhile, things have been both heating up and cooling down in the world of HPC. Larger scale-out systems and faster interconnects have pushed usable core counts higher, while cooler and more energy efficient processors, vector architectures, GPGPUs, consolidation in the cloud, etc., led the way to more diverse and sustainable HPC hardware ecosystems.
Modern and upcoming HPC systems present a truly mixed bag of performance and efficiency choices that require careful re-tuning of software components. From hybrid programming for shared/distributed clusters dawned the era of heterogeneous computation including accelerators such as GPGPUs and Intel Xeon Phi, and associated parallel patterns and programming languages. Continuing to enhance scalability and throughput in this new computational environment in order to meet the increasing demands of industry to support their simulation-driven product development processes is the challenge of the coming years.