Impressive Scale-up Performance with Azore® CFD
By Kelly Hile
Computing resources today are improving at a rapid pace. High performance computers have more cores than ever before, and the numbers keep climbing. In 2023, some double processor CPUs are up to 196 cores on a single machine.
With core counts skyrocketing, it’s great news that Azore has licensing options with unlimited cores, enabling users to put every single core they own to work, not restricted by complicated HPC licensing schemes. Our licenses even extend with no fuss to cluster computing, so you can run in parallel with other cores on your network.
Apart from the licensing freedom to actually use all of your cores, the real question is one of performance. How does Azore utilize extra cores and what affect does that have on run time? We quantify this characteristic as speedup.
Let’s say you are running a solver with a single core, on a single machine. If you run the same simulation and engage two cores, your simulation will be faster. The extent of this speed increase is the speedup. For each core added, you want to be able to get faster simulation time from your solver, and Azore delivers on scaleup potential.
In our Azore tests, we found high scaleup efficiency with a 60% or higher speed increase for each additional core.
This level of efficiency kept up whether the combined cores were running on the same machine (OpenMP) or on a network (MPI). Azore ensures parallel runs don’t add any steps to your workflow, either. Running on multiple cores is just as easy as running on a single core.
In one benchmark test, we ran a 108 million cell model with a size of about 100 GB of RAM that featured external flow, energy transport and species transport.
We used a basic Beowolf cluster with two networked CPUs (2 nodes). Each node is equipped with 20 available cores, for a total of 40 available when running in MPI mode. Both machines contain Intel Xeon E5-2690 V2 processors.
We measured the time it took to complete a set number of iterations for an increasing number of engaged cores, from 1 core to all 40 of the available cores on this cluster. Here’s how Azore performed:
Azore not only performs faster with each additional core, but also performs faster on a cluster than with a single node for the same number of cores. Azore’s scale-up performance was consistent in a number of other benchmarks, performed with additional CFD models and clusters, up to 108 cores and 3 nodes. Download the white paper for a summary of the cluster configurations and solver performance data.
These real-world cluster benchmarks give users assurance that investing in additional cores and processors will have a real impact on simulation time. When a software package combines excellent scale-up performance with no fuss licensing, the sky’s the limit.