Adding Nvidia Maximus software and a Tesla GPU to their workstations gave 2x boost in Ansys and MSC simulations with up to 8 million degrees of freedom.
Parametric Solutions Inc. (PSi) is an engineering services company that performs mechanical design and analysis work for a wide variety of clients, from solo inventors to large corporations. The industries supported include aerospace, consumer products, medical equipment, green energy, and transportation. Since 1993, PSi has tackled projects throughout the full product development cycle, from initial concept to manufacturing. PSi relies on state-of-the-art design and engineering skills it has honed in its main field of specialization: the power generation and industrial/aviation gas turbine industry.
Much of PSi’s work entails complex analysis and design: thermal, structural, fluid dynamics, vibration, kinematic synthesis and optimization, and the coupling of these disciplines in fully integrated simulations. PSi’s team of more than 70 engineers, many of whom use Ansys software for their work, have realized a remarkable boost in performance and productivity with new Nvidia Maximus technology, which lets them run Ansys compute-intensive processes while continuing to do other work, such as CAD modeling, simultaneously on a single workstation.
“Nvidia Maximus technology is an extremely powerful tool for the way we work. With the 2x performance boost we saw immediately, our new system has completely changed our workflow and productivity,” said David Cusano, chief engineer and co-owner of PSi.
8 million degrees of freedom
PSi works with big simulations. For instance, a structural analysis model for a large turbine blade might have between 4 million and 8 million degrees of freedom (DOF). Even a small, “quick” simulation might require 2 million DOF.
“Before we got the Maximus system, the solving requirements for a typical simulation are such that we couldn’t do design and analysis on the same box,” said Cusano. “The Ansys analyses, in particular, require such large computing horsepower that it would take days to run a single analysis if we wanted to work on the box at the same time—and even running it on its own involved tying up the machine for 8 to 12 hours or more.”
Their solution was to set up a separate run box—a quad-core CPU with eight 3 GHz processors, 16 GB of RAM, and zero disk array for storage—to do the heavy computational work. No one would sit in front of the run box, and engineers had to send work remotely or physically walk over to the box to initiate analyses, and then return to a separate computer to do other work.
“This process was far from ideal,” said Cusano. “Transferring files over the network back and forth is a hassle, and it was inconvenient to interact with the runs or check them as they went along. But the biggest problem was that it was a shared resource, so engineers could never be sure of when they’d have access to it. If they needed rapid turnaround between design and analysis, they could be out of luck.”
PSi installed a Maximus configuration, consisting of a Nvidia Quadro 6000 GPU alongside a Nvidia Tesla C2075 companion processor, all running on a Dell Precision T7500 Tower Workstation. Typically, a PSi engineer’s workflow starts with conceptual design of a component in CAD modeling software such as Creo (formerly Pro/Engineer), NX, Catia, or SolidWorks. The model must then be analyzed in detail to evaluate the materials, durability, and other requirements for that component. PSi uses Ansys structural, thermal, CFD, and kinematics modules, as well as similar MSC products.
“The Maximus system allows us to combine those previously separate activities—CAD modeling and compute-intensive analysis—into a single entity on the same box,” said Cusano. “In addition, the performance boost allows our engineers to have a resource that can keep up with the way they need to work.”
For example, a PSi engineer working on an industrial power turbine module, such as the large blades that attach to a huge steel rotor, needs to perform various thermal, structural and computational fluid dynamics (CFD) analyses in ANSYS – then feed back the results of this work into the design of the parts. This process is iterative and time-consuming, especially when aspects of the process require moving the analysis to different resources, and waiting for the results.
“That engineer can now do all those analyses on his own workstation using the GPU instead of the CPU, about twice as fast as before, and work on the next step of the process on the computer at the same time,” said Cusano.
“When you can run everything you need on your own computer, you’re never short of resources like you are when everyone’s forced to share the same run box,” said Cusano. Previously, PSi’s fastest dedicated run box could solve a job in about 10 minutes per iteration. In a typical 50-iteration process, that’s 500 minutes of time—or more than 8 hours—where the box would be tied up doing a single analysis.
“With the Maximus system, we can run that same process using a much less-expensive workstation [and] that same ‘typical’ workstation can still be used for other jobs at the same time,” said Cusano. “It’s amazing not having to support the dedicated run boxes, separate from the rest of our tasks. It changes the way we work as a company.”
It also translates into cost savings. By using Nvidia Maximus-equipped workstations, PSi saves the direct costs of not having to use the expensive dedicated run boxes, as well as now getting more productivity from each engineer’s existing workstation.
“I would never want to go back to pre-Maximus days,” said Cusano. “It would be too much of a performance hit. Doubling the speed of the box for analysis, plus the ability to do other applications in parallel, provides us direct and impressive productivity gains, which is core to PSi’s business.”