You have no idea how easy and cheap it is to have access to high-performance supercomputers for your CFD simulations with SIEMENS STAR CCM+... Whether you are a large corporation or institution, whether you are a startup or industry newcomer to the world of CFD. Let's delve into this topic in a practical and didactic way. Follow:
I'm Ricardo Damian , Technical Director at CAEXPERTS, a computer simulation enthusiast, with more than 25 years of experience in CAE, always helping industries to implement simulation technologies to solve industrial problems.
In the past, it was not easy to use CFD, use high-performance computing, you had to know UNIX, LINUX, vector programming, parallel programming, scientific computing architecture, a lot of dedication and patience. Windows was not an operating system for scientific computing.
I started basic training in Mechanical Engineering at UFRGS, already with an emphasis on CFD, at the Super Computing Center at UFRGS. Then I did a sandwich degree at the University of Stuttgart in Germany, I did an internship at the BOSCH Research Center there in Germany and then at EMBRAER here in Brazil on the way back. Then I did a master's degree in Chemical Engineering at COPPE/UFRJ, with a great class in HPC. Using CFD at the time was tough and running on a cluster was a privilege that required a lot of preparation.
Around the 2000s, the first PCs with more than one processing core began to appear and the CFD codes already ran on Windows, but with lower performance. During the next 10 years, I witnessed the beginning of the use of parallel processing clusters in Brazilian universities and research centers. Only a few large industries, with a mature CFD usage , had access to HPC clusters . I worked at PETROBRAS CENPES supporting more than 200 CFD users at the time to use the clusters , installing, configuring, specifying, administering, optimizing performance, etc. It was a super technical activity that involved several professionals. From 2010 onwards Windows started to be good for HPC, but Linux has always reigned in the subject and still reigns today.
Nowadays, virtually all CFD simulations are performed on more than 1 processing core. We have available CPUs with 4, 8, 10, 16, 20 processing cores, being able to operate with 8, 16, 32, 64 or even 128 Gb of RAM. Furthermore, state-of-the -art CFD codes, such as SIEMENS Simcenter STAR CCM+, already allow the use of new technologies such as GPU or ARM architecture processing.
In other words, today's Workstations already allow for high calculation performance, which makes complex CFD simulations available to novice users. I myself, in my daily use of STAR CCM+, prepare the simulations and run many of them on my personal and work laptop, with an 8-core i7 processor and 16 Gb of RAM.
I learned early on how to simplify simulations well. I really like to start CFD studies with simplified simulations, with a coarse mesh, only with the physical models really necessary, building confidence in the model, testing the sensitivity of the parameters (geometric, operational, properties of the fluids/particles involved), to then, if necessary, go to HPC. I often manage to arrive at the final solution without even going to a cluster . That's one of my skills. I am an exception, as nowadays 80% of CFD simulations in the world take place in clusters .
Einstein once said that we should try to make things as simple as possible, but not simpler. I am fully aware that there are many cases where CFD needs to go to heavy parallel computation, transient studies, detailed turbulence (LES, DES, DNS), multiphase problems, with particles (DEM), phase shift, flow regime transition, shape optimization, parametric optimization, experiment design, robust design, population balance, moving mesh, vortex-induced vibration, aeroacoustics, complex reactions with tens or hundreds of components, etc.
Having a physical cluster in your company means an equity investment in the order of millions of reais and highly perishable (with 2 years it starts to have hardware problems , with 4 years it is outdated and with 6-8 years it is better to turn it off because it consumes a lot of energy) , and you spend practically the same amount on a CPD infrastructure with air conditioning , IT security, software (OS, administration, etc.) availability.
Then the provision of servers in the cloud began to emerge, also for scientific computing, and the thing became more professional for CFD as well and finally, today we have a wide range of choices of companies that rent cluster services in the most diverse ways. Recently, SIEMENS launched Simcenter Cloud HPC, initially for STAR CCM+, which really is a disruptive technology.
Why disruptive? You see, now you can buy prepaid credits for cluster hours, with machines from 141 to up to 756 cores, which are really cheap (since shared use in the cloud on a large scale lowers costs), and start using CFD on cloud already. In addition to our internal use, we are serving companies of all sizes, which are already starting to use CFD with access to HPC in the cloud, from startups to consolidated industries. High performance CFD has never been simpler to use and more affordable in terms of initial investment.
For example: In a CFD simulation of a complex process, such as the casting of complex geometry parts, where a good detailed study requires a mesh of about 5 million elements, long transient, adaptive time step, several phases ( liquid and solid metal, air, sand and porous filter), interface monitoring, conjugated heat transfer (conduction, convection and radiation) and backfill flow control. Normally, in a good Workstation with 16 cores and 64 Gb of RAM, it would take about 16 hours to solve the whole process flow (filling, solidification and cooling). If we take it to the cloud, the simulation is ready in 30 minutes. It changes everything!
Source: Aberdeen Group, Cloud accelerates digital transformation blog
We can quickly do some calculations of return on direct investment, considering the cost of software license , hardware rental , depreciation, IT administration, and the main cost of the product development engineer. And I'm already advancing, it's really worth using CFD in the cloud, looking at it from any point of view. But what about overhead costs?
What is the cost of waiting weeks to design a mold? What is the cost of not testing your product or process a lot? What is the cost of not knowing in depth the behavior of your product, equipment, process, plant? What is the cost of not testing the most critical scenarios first? What is the cost of having to release a product that has not yet been thoroughly tested? What is the cost of not closing a deal because your product is using too much raw material and energy? What is the cost of being number two? Or the competition coming out with something better before you? What is the cost of a RECALL?
Incredible as it may seem, many companies are still stuck with the concept of HAVING fixed assets. Usually industries invest heavily and in the long term in the purchase of machines to manufacture their products. Investments in product engineering, process, research, development and innovation tend to be lower. When it comes to investing in renting software and hardware in the cloud, many objections arise. Digital technologies are in full evolution and contracting the As-A-Service type is very advantageous. What good is an efficient typewriter factory?
See in this video how easy it is to perform a cloud simulation with STAR CCM+.
I hope I helped break some paradigms and motivate your company to learn more about how to make complex CFD simulations simpler and more available.
Talk to CAEXPERTS and understand why we are experts when it comes to helping your company accelerate innovation and competitiveness with digital engineering!
Comments