We are so thankful to our awesome customers! This post is a special “thank you” to all of you who attend our webinars and ask such excellent questions. It is through your questions that we learn what’s important to OnScale users and this, in turn, allows us to continuously improve our software. The conversations that start in our webinar Q&A often carry over into the OnScale offices and we’re always happy when we hear them coming up in our development meetings.
Thank you for being here for OnScale! For additional questions please reach out to email@example.com.
Q: What is a core-hour?
This is a great one, as it’s so fundamental to how simulations are billed on OnScale’s platform. The core-hour (CH) is a measure of computational effort. It’s the product of the number of cores used for a simulation and the number of hours the simulation takes to run. So for example:
A simulation running on 8 cores for 1 hour uses 8 CH.
A simulation running on 1,000 cores for 0.1 hours (6 minutes) uses 100 CH.
OnScale free account users get 10 CH every month, enough for 5 hours of simulation on a 2 core cloud node – plenty of opportunity to try out OnScale’s capabilities.
Q: How long do typical simulations take?
This is one of the most common questions we get. To give the best answer possible we thought it would be helpful to answer via an example.
Take the PMUT 2D from our advanced examples. This is a time domain problem, where we simulate the performance of a simple ultrasonic sensor. While the design is simple it still requires 42972 degrees of freedom to solve which is pretty large by traditional FEA standards. At each of the 92300 timesteps, we are solving the mechanical displacements and electrical displacements at each node. Despite this OnScale’s solvers complete the simulation in 50 seconds on a 2 core node. That’s only 0.027 CH
At the other end of the spectrum we solve big problems involving billions of degrees of freedom, which can take a few hours to compute.
So in truth, simulation time depends on the problem, but it might be shorter than you think. The best way to get an idea of how long your problem will take is to check out our advanced examples.
Q: Do simulations run in parallel on OnScale – how does that work?
The short answer to this one is yes, they do! Traditionally the barrier to running engineering simulations in parallel has been licensing. For example, if you buy 1 license to a legacy package, you are only allowed to run 1 simulation at any given time. Even if you bought many licenses (which is very expensive), you would still be limited by the local hardware you have access to – at some point you would run out of cores.
We think that’s a huge bottleneck, and it’s something we set out to fix with OnScale.
With OnScale you have an unlimited number of licenses. You can run literally, 1,000s of simulation in parallel. Tomorrow you could invite a colleague to your account, and you could both run 1,000s of simulations in parallel. There’s no limit!
What’s more, OnScale’s cloud hardware is completely scalable, so you’ll always have access to the number of cores you need to run your job. Need 16 cores for each of those 1,000 simulations? Sure, no problem. 16,000 cores spool up for you on demand.
Q: Can I use more cores to make jobs run more quickly?
OnScale’s solvers have been designed from the outset to be parallelized across many cores. This means that they can take advantage of highly scalable cloud hardware to deliver problems very quickly.
What’s interesting is that we’re not limited to simply spooling up the largest node that a cloud provider has to offer, often around 128 or 256 cores. We can actually go further than that by creating a High Performance Computer (HPC) on the cloud, which is made of 1,000s of separate cloud machines. We call this Cloud HPC. This provides OnScale customers with HPC level power on demand, allowing them to accelerate their models and get results quickly.
Q: What meshing should I use? How will it affect accuracy?
This is a universal question that engineers ask when looking at a numerical simulation tool. We all know meshing is important but finding the right mesh for a particular problem is sometimes a grey area.
The great thing about OnScale is that the ability to run simulations in parallel makes it very easy to analyze what mesh settings are right for your needs. The approach we take is to pick an output that is important to you (it could be the amplitude of a signal, or the frequency of a resonance) and see how that varies with meshing. That way you can make an informed engineering decision on what mesh is right for you.
Don’t miss our next webinar on Wednesday 7th August at 9am (PDT) to learn Why Core-Hours make Cents! We encourage lively discussions in our webinars, so stay tuned to see if your questions feature in a future blog post.