Altair’s structural analysis and industry leading optimization software, OptiStruct, has become much faster in the last three years due more efficient algorithms, as well as an increase in calculation speed due to hardware improvements. Altair worked with Dell and Intel to benchmark a three year old version of OptiStruct on a three year old Dell computer and Intel chipset, against a current version on OptiStruct running on new hardware. In just three years, speedup from factors of 8 to 18, depending on the analysis type, are obtained.
Faster Hardware and New Algorithms run 8 to 18 Times Faster
In recent years, the number of cores on a computer chip has increased dramatically. A typical deskside workstation can have 16 cores now, which allows more computations to be run in parallel, leading to shorter run times. In addition, the decrease in the cost of Solid State Drives (SSD) allows them to be used more often for storing data. The decrease in the price of computer memory (RAM) leads to the use of more memory for in core calculations and I/O file caching. Finally, and most importantly, new algorithms developed by Altair to take advantage of more cores have significantly reduced the times needed for structural analysis.
The older computer had just eight cores and a single mechanical hard drive. The newer computer had 16 cores and two Solid State Drives (SSD) in RAID0 configuration. The RAID0 configuration allows for reading or writing to multiple drives at the same time. This simultaneous reading and writing can nearly double the data transfer rate. In addition, the older computer had 128 GB of RAM and the newer computer had 198 GB of RAM. More RAM means less data to be transferred to disk, as well as providing more RAM for data I/O cache.
Results for Different Analysis Types
Dynamic condensation is used to generate a crankshaft model for use in a multibody dynamic analysis. In this case, we took a 12.2 million Degrees of Freedom (DOF) model and reduced it to 2,226 connection DOF and 884 normal modes. With old software on old hardware this took 921 minutes, while in new hardware it took just 493 minutes for a speedup of nearly two times. Using a new algorithm for dynamic condensation on the new hardware, reduced the calculation time to just 111 minutes for a speedup of over four times. The total speedup was over a factor of eight.
If we look at the static analysis of an automobile steering knuckle modeled using 15.2 million DOF, the analysis time with the old software on the old hardware was 803 minutes. Using the old software on the new hardware it was just 369 minutes, for a speedup of over two times. Using a new static analysis algorithm on the new hardware, the analysis time was reduced to just 68 minutes for a total speedup factor of nearly 12.
Finally, we look at the nonlinear contact analysis of a rail car modelled with 5.4 million DOF. There are over 10,000 contact pairs in this model. On the old hardware with the old software the analysis time was 674 minutes, while it took just 284 minutes on the new hardware. The new hardware delivers a speedup of about two and a half. Using a new contact algorithm, the new software runs in just 38 minutes on the new hardware for a speed up of over seven. The combined speedup using new hardware and new software is nearly 18.
The utilization of new hardware delivers a speedup factor of over two comparted to just three years old hardware. New algorithms in OptiStruct provide an additional speed up of between five and eight times depending on the analysis type. Reduced analysis time leads to the ability to make multiple design changes in a single day. In addition, some of the analysis times are so low that Monte Carlo simulations can be run in order to check the reliability and robustness of the design.
Read more about the benchmark result on Desktop Engineering.