By Liam McManus
Unintentionally, while trotting several kilometers of the Thames Path in preparation for the marathon, I found a piece of Siemens history right outside my door. Southeast London’s Woolwich Dockyard neighborhood was home to a sizable facility owned by the Siemens Brothers Co. Ltd. of Woolwich, London. What intrigued me even more was that, among many other things, this facility produced dry cell batteries in the 1920s! Radios and automobiles were the main uses for these batteries. Why is this so fascinating to me? If we jump ahead a century, the focus would be on batteries in the car industry.
Naturally, the goal is to power the vehicle’s propulsion system rather than the radio or spark plug like those outdated Siemens Brothers batteries did. Fortunately, at least one Siemens salesperson in the Royal Borough of Greenwich, as well as hundreds more across the globe, are still willing to assist!
Electric vehicle powertrain designers can expedite all areas of powertrain design by using Simcenter STAR-CCM+ to explore thorough Cell Design, Thermal Runaway, and Thermal Management of battery packs with a single tool.
The need to boost simulation throughput within constrained timeframes is driving an increasing demand on engineers. An increasingly crucial component of the jigsaw for achieving this kind of aim is GPU hardware. I looked at the amount of publications at ScienceDirect that mentioned “GPU accelerated CFD” out of curiosity. Although the method may not be strictly scientific, the findings suggest that interest in the field of GPU-powered CFD simulations is expanding quickly.
We have made even more progress in Simcenter STAR-CCM+ 2402, extending into detailed thermal management, in terms of both the physics that can be natively represented on GPUs and the spectrum of GPU hardware that can be employed! Here’s a recap of some of the programs that GPUs equipped with Simcenter STAR-CCM+ can run natively.
Let’s start with the hardware and go from Southeast London to AMD’s headquarters in Silicon Valley. Why? Since AMD GPUs can now be used for GPU-native computation in Simcenter STAR-CCM+ 2402! This signifies a significant advancement in expanding the range of GPU hardware that is accessible and usable. We especially suggest the AMD InstinctTM MI200 series, which includes the MI210, MI250, and MI250X. Additionally supported are the RadeonTM PRO W6800, Radeon PRO V620, and Instinct MI100. Simcenter STAR-CCM+ is now the only multi-purpose CFD code that is compatible with the entire range of CPUs and GPUs, which puts it in a unique position. This is an additional step in the process of rethinking standard CFD design approaches through GPU architecture, as I have discussed previously.
We are eager to collaborate with AMD to further broaden the range of hardware options in upcoming releases (such the Instinct MI300 series) and to keep searching for ways to improve the performance of CFD simulations running on GPUs.
Using a Conjugate Heat Heat Transfer (CHT) methodology, I choose to simulate the thermal behavior of a battery pack, armed with three AMD MI210 GPUs and my memento of local Siemens history. Fortunately for me, Simcenter STAR-CCM+ 2402’s recently added GPU-native capabilities made this achievable.
The Multi-Component Solid model comes first. This makes it simple to apply various material qualities to intricate assemblies made up of several pieces. I can therefore simply describe several materials for busbars and positive and negative terminal connectors in this example under a single physics continuum. Even though we are only using two materials in this battery example—copper and aluminum—full vehicle thermal simulations can specify many more materials. Moreover, having multi-component solid capability built into GPUs is another step toward leveraging the advantages of GPUs over traditional CPU architectures in terms of speed, cost, and performance.
Using Mapped Contact Interfaces (both implicit and explicit) in CFD simulations with GPUs is the second GPU-native feature added to Simcenter STAR-CCM+ 2402. Once more, this is a crucial facilitator for thermal management modeling, particularly in the context of varying timelines.
[/vc_column]
Upon comparing AMD EPYC 7532 and AMD MI210 CPUs, it is evident that the utilization of GPUs yields performance comparable to hundreds of CPU cores. Additionally, we can demonstrate that employing GPU hardware can dramatically minimize energy overheads while running these simulations by estimating the energy usage for each setup (using given TDP values).
AMD EPYC 7532 dual socket CPUs versus AMD MI210 GPUs in terms of runtime and energy usage. The TDP values of 400W for each EPYC 7532 dual socket (64 cores) and 300W for each MI210 were used to estimate the energy usage.
For example, the 16kWh saved here could fuel a battery electric vehicle (BEV) for 80 km on average. After a year, the energy savings would be sufficient to power your BEV all across the world with extra energy left over.
We acknowledge that there is no magic bullet when it comes to conducting your CFD simulations, even though these data show the potential that using GPUs to execute CFD simulations offers. Having said that, HPC clusters equipped with respectable CPU technology will still have room and will continue to be a significant computational powerhouse. Because of this, our approach supports a broad variety of compute devices and guarantees a smooth transition between them. This includes knowing that whether you use GPUs or CPUs to complete your task, you can be sure to get consistent results.
The HPC scene for computer-aided engineering applications has evolved over time, much like Southeast London’s post-industrial landscape did. The ability to genuinely boost your simulations by cutting turnaround times, operating expenses, and hardware prices is made possible by the use of new hardware architectures to CFD. In the end, this translates into improved goods, more designs, and higher simulation throughput. It’s time to consider what engineering problems Siemens will be solving in a century. In any case, I’m sure that simulation will take center stage!
All you have to do going forward is continue to feed your CFD-hungry (AMD) GPUs for a little while longer. Until February 28, when Simcenter STAR-CCM+ 2402 comes, then the great feast starts.
