High Performance Computing (HPC): With Great Power Comes Great Availability
When I started my career in simulation, having high performance computing was a costly endeavor. Having 64 CPU cores to run a CFD simulation job was considered “a lot”, and anything over 128 CPU cores meant you could run through giant simulations with great speed (for the time) and was typically the purview of research facilities or large corporations.
But today, a common workstation laptop can have 32 CPU cores, and simulation engineers are talking about running simulations on hundreds of cores in day-to-day conversation. And, with the advent of GPU computing cutting runtimes down significantly, HPC is becoming a common phrase across the wider industry.
But is it attainable for a smaller company just starting their simulation journey, or is it just for big companies? The real question is, can you afford not to use HPC?
You Can Only Do So Much with Four Cores
Let me open with the statement that you can already do a lot of simulation on just four CPU cores. A lot of customers will be comfortable with running simulations at this quantity (standard for Ansys Fluent and Mechanical).
READ MORE: Simpler, Smaller and Smarter: Simulation Tactics for Better Product Design
However, there are some products and projects that require more horsepower. So, when someone is designing a complex product that requires more than just four CPU cores for simulation, one of three things occur:
- Lots of prototype testing, which incurs time, cost and rework
- Leveraging simulation to run several design iterations
- A blend of both
A customer obviously wants to reduce physical prototype costs and tooling development delays, which is probably why simulation was deployed in the first place. But now envision having two weeks to run simulations on a new product with only a four CPU core workstation. This hypothetical product will likely have 12 different operating conditions, and you only have the computational resources to run five of those conditions.
Which run conditions do you choose to simulate? The best- and worst-case scenarios are obvious choices, but what about the remaining three cases to simulate? The take-home point: Product design performance can be hampered without the proper computational resources.
READ MORE: Simulation Aids in Product Development and Design Across Industries
Now, in that same scenario, envision having HPC resources that give you access to 32 CPU cores on a workstation. In the same two-week timespan, it may be that you now can not only run all 12 operating conditions, but you can also simulate design improvements on an additional 12 operating conditions.
Perhaps this avoids retests, project delays caused by last-minute redesigns and warranty claims. HPC allows you to have both the quality and quantity your design needs to be robust in customers’ hands.
HPC Doesn’t Require Hundreds of Cores to Be Effective
The next myth you may have heard was HPC requires a huge investment or hundreds of CPU cores to be useful. While some customers may in fact require hundreds of CPU cores due to the complexity of their models, many newer users don’t necessarily need that level of financial investment.
It can be as simple as moving from a 4 or 6 CPU core laptop to a workstation with 16 or even 32 cores to start seeing a noticeable increase in computing speed. No complex infrastructure, no off-site computing.
If a high-end workstation investment is cause for concern, compare the cost(s) associated with a delayed product launch or failed physical tests. If there is a repeating pattern of excess retooling costs, weeks of extra design rework or frequent launch delays, the cost of a new workstation could offset or even eliminate a lot of those recurring costs.
READ MORE: Crunching the Numerics: Computer Aided Engineering
Another alternative to more CPU cores is utilizing a modern GPU for simulation computation. GPU computing can be used in a variety of Ansys tools such as Ansys Discovery and flagship tools like Ansys Fluent. Many benchmarks have shown that GPU computing can outpace traditional CPU cores in a variety of simulation applications, some even showing speed exceeding four times that of equivalent CPU cores.
You may find that it’s an even cheaper route to buy a GPU and the associated software licensing than building a custom workstation with equivalent CPU cores. If you’re curious, read more about our GPU benchmarking study.
Times are changing fast, and so is the rate at which simulation can solve your design problems. There are HPC resources available for all kinds of budgets and the reason to “get by” with less in simulation is becoming obsolete.
