This week ANSYS announced revision 12.0 to its suite of physics simulation software. The company is emphasizing HPC capabilities with this release, and users can expect improved performance on everything from multicore, single-socket workstations to mid-scale HPC clusters of a couple thousand cores. We talked with Barbara Hutchings, the director of strategic partnerships at ANSYS, and Ray Browell, senior product manager, about the new release, and about the company’s position in the HPC ecosystem.

Founded in 1970 and now a public company employing about 1,700 employees, ANSYS is all about engineering simulation software. From its beginnings in computational mechanics, the company’s software now covers everything from fluids to electromagnetic simulation plus the pre- and post-processing ANSYS users need to prepare problems for solution and to analyze the results. Some of these codes, like FLUENT and LS-DYNA, are long-time staples of the high-performance technical computing ecosystem. ANSYS customers are in virtually every industry — from power and biotechnology to consumer products, automotive, aerospace and electronics. According to the company’s Web site, 18 of the top 20 innovative companies in the world (as identified by BusinessWeek magazine) are ANSYS customers. The companies on that list are household names: Apple, IBM, Sony, Nokia and others.

ANSYS 12.0 bundles improvements across the entire ANSYS suite of products, all aimed at responding to the growing ubiquity of multicore desktops and cluster computers. As Hutchings puts it, “HPC is a core technology for ANSYS because it is very important for our customers. Simulations that have to be done in time to impact product engineering decisions need HPC.” She says that ANSYS customers today are split between those that run simulations on high-end desktops, and those that run on clusters ranging from tens to thousands of cores.

Starting at the most basic building block, ANSYS 12.0 improves multicore performance with an emphasis on both Intel and AMD processors. According to performance figures from the company, the new software on an eight-core Windows system can deliver a speedup of five to six times in mechanics simulations, and seven times in fluids simulations. Some of this is due to the improved memory bandwidth of the new Nehalem processors, but the company has also invested significantly in new algorithms, better cache (re)use, and better domain decomposition and load balancing techniques.

At a larger scale, ANSYS 12.0 offers good scaling behavior for structural mechanics simulations up to 512 cores (depending upon the problem and the solver being used). This improvement is significant given the typically limited scaling behavior of these types of simulations. Even fluids problems, which typically scale more effectively, experience better performance, with nearly 80 percent of ideal speedup up to 2k cores. Better domain decomposition and load balancing techniques contribute to improved performance here as well, as do improved communications algorithms and new solvers that are more effective at scale.

And thanks primarily to changes in the underlying software architecture, ANSYS products are now able to support simulations with over a billion degrees of freedom. Hutchings says that this is about an order of magnitude larger than the typical high-end application in her installed base, but pioneering customers in industries like automotive and aerospace are already pushing those limits as they expand simulations to include more detail and model entire systems.

Another area of significant enhancements that HPC users will be interested in is file handling. Improvements include support for parallel file systems using MPI-IO, and the company’s benchmarks demonstrate (for example) a 20x improvement on file write performance for a 256 core solution using ANSYS FLUENT on a Panasas file system (compared with serial I/O). Other improvements in I/O include file compression when gathering results from cluster nodes in ANSYS CFX simulations.

The ANSYS 12.0 suite runs on Red Hat and Novell Linux distros as well as Windows HPC Server 2008, and is Intel Cluster Ready. ANSYS has also teamed up with HPC hardware vendors like HP, SGI, Cray and others to define specific configurations designed to run their software most effectively.

What about the future? One of the things the company is actively investigating is the use of GPUs for faster solutions. According to Browell, this is an area of “active research,” but they aren’t ready to announce support just yet. One of the stumbling blocks is that ANSYS simulations rely heavily upon fast double-precision math, and while GPUs are continuing to improve on this front, their strength is still single-precision performance. But this is rapidly changing, and ANSYS is positioning itself to be ready as GPU chips continue to evolve.