In a recent blog post, Wim Slagter, lead product manager for High Performance Computing at ANSYS, Inc., tackles a set of six myths of high performance computing gleaned from conversations at various ANSYS Convergence Conferences.
These six myths or misconceptions deal with HPC as it relates to engineering simulation:
+ HPC is available on supercomputers only
+ HPC is only useful for CFD simulations
+ I don’t need HPC – my job is running fast enough
+ Without internal IT support, HPC cluster adoption is undoable
+ Parallel scalability is all about the same, right?
+ HPC software and hardware are relative expensive
Slagter addresses the first three, promising to revisit the latter three in a separate entry.
The first one really depends on your definition of HPC. If you have a hardware-based definition, then you may agree with the idea that HPC is mainly the purview of supercomputing or big iron. It’s Slagter’s take, however, that HPC has evolved away from that narrow definition.
“A decade ago, HPC may have indeed been primarily associated with big supercomputers. However, the computer industry has delivered enormous increases in computing speed and power at consistently lower costs,” he writes.
“Think about more compute cores per CPU, integrated I/O on processor die (yielding higher memory bandwidth), more and faster memory (channels), larger L3 cache size, faster disk storage (like solid-state drives for ANSYS Mechanical), faster interconnects, AVX support, etc. Through these advances made I can counter this myth #1 by stating that HPC is today available throughout the entire computer spectrum, at the entry level from tablets running ANSYS Mechanical to multi-core laptop and desktop computers, as well as the bigger workstations but of course also computer clusters at the other end of the computer spectrum.”
As for HPC being a CFD-only thing, Ansys has a full set of structural mechanics benchmarks – covering different solver types and model sizes – showing scaling to 80 cores. There are similar benchmarks for electromagnetics simulations. So while CFD may scale higher, that doesn’t mean that other codes don’t also benefit from parallel processing and HPC tools.
The third one is perhaps the most interesting… the contention that HPC is unnecessary.
Perhaps there’s a lesson from the classic quote attributed to Henry Ford: “If I’d asked people what they wanted, they would have asked for a faster horse.”
Slagter’s response to an FEA user’s claim that the job runs fast enough is to share about the benefits of high-performance computing using real performance data targeted at different solution areas.
As just one example taken from a 2013 IDC report, results indicate that for ever dollar invested in HPC, $356 dollars gained, $38 dollars of which is seen in profit.
There’s a lot more material to check out at http://www.ansys-blog.com/six-myths-of-high-performance-computing/.