ANSYS claims that this efficiency in itself is worth noting as competitor simulation packages tend to operate at around 75 percent efficiency when approaching this scale.
It wasn’t long ago that ANSYS announced that Fluent was scaled up to 36,000 cores. ANSYS has since increased the scope of their scalable simulations capabilities.
"This [3.6x] increase over the previous record for compute core count includes a complex simulation involving multiple complicated physics, in this case an industrial combustor," said Wim Slagter, ANSYS HPC product manager. "We test our software on real-life applications by partnering with the global leaders in supercomputing. That's the only way we can ensure that the advancements made in the lab are actually usable by our industrial customers."
This breakthrough was made with the help of the National Center for Supercomputing Applications (NCSA), the National Energy Research Scientific Computing Center (NERSC), and supercomputing company Cray Inc.
"Less than a year ago, our partners ANSYS and NCSA made a breakthrough with a 36,000-core fluid dynamics simulation," said Steve Scott, chief technology officer at Cray. "Only a few months later, the team has eclipsed that record. This example of extreme scalability for high-fidelity simulations is needed to keep up with the accelerated pace of innovation in the marketplace."
By increasing the number of cores ANSYS can work with, customers will be able to run more complex simulations faster, allowing for more iterations to optimize designs. Though most customers will not have access to computing power near this record, they will still be able to improve the speed of their simulations on smaller HPC setups thanks to this research.
Within the last decade, it wasn’t uncommon to hear that the maximum scalability for a simulation was around 512 cores. However, ANSYS has been able to double their scalability annually.
Another simulation company, CD-adapco, has also improved their scalability, setting records of 102,000 cores and 55,000 cores, making what is turning out to be a significant rivalry in the world of simulation HPC.
How many cores do you have to work with? Do you see a point in using over a quarter of a million cores? Who do you think will dominate the HPC competition? Comment below.