So, it’s time to talk about CFD in 2014.
Of course, I don’t want to pretend to be an industrial analyst for CFD. I’m not rubbing the crystal ball. Instead, I’m ranting from the perspective of end users, because I think I am part of them.
Technology is abused nowadays.
If you ever watched a shopping channel (maybe just accidentally in a hotel), you may notice “technology” is one of the hot words in the show. Maybe just for a mop, or an egg mixer, the presenter is always intoxicated by the “technology” the product has. The “technology” talked may be just small tricks or common-sense.
If you think CFD software industry is an exception, you are probably wrong.
The technology behind CFD is not evolved as fast as CFD software releases. But most CFD software vendors need to push incremental update regularly so that the customer can keep spending.
One way to justify the upgrade is some sorts of new technology. It is not strange to hear more and more “technology” in the CFD software releases. Big vendors do this, so do small vendors.
But if you check such claimed new technologies, most of the time, you may find out they are just another tricks or minor improvement.
This is not the fault of CFD software vendors. There are simply not enough excitements in CFD technology.
The harsh fact is that more and more CFD software now shares the similar level of technology. The judgement of CFD software technology becomes more and more subjective. So, it is now very easy to bash other CFD software competitors.
The battlefield among CFD software vendors is gradually being shifted to branding. Technology is still important. But it is not as important as it used to be.
Of course, branding is always important. But in the old days, vendors usually brand themselves through distinct features or underlying technology or particular applications. This works well when users know what they want and understanding how the work should be done.
As technology in CFD evolves slower than software releases, all CFD software vendors, especially traditionally small ones, have time to extend the capability chart. It’s not surprising most CFD software has similar capability chart. No CFD software vendors are relying on some unique features or unique capabilities nowadays.
Another driving force for branding is from users. More and more CFD software users are CFD guerrillas (I talked this a bit in here and here). They are not dedicated CFD analysts. Most of the time, they are mechanical designers/engineers. Understanding how the CFD software solves their problem is not in their job scope.
This is not a question of good or bad. This is the trend.
Branding is not bragging. The CFD software vendors now need to not only understand customer’s problems, but also educate them to use their software to solve the specific problem in a correct and efficient way. Building trust with potential users is more and more critical for CFD software vendors.
When CFD software expands its capabilities, it becomes more and more sophisticated. All vendors want to pack more and more. This creates an unnecessary cognitive load on the users. The context menu may alleviate this problem to a certain extent, but will not solve it.
And the truth is that more than 99% of users will use less than 1% features or functions the CFD software provides.
So providing tools or products for specific problems is one of the possible approaches.
In the market, some CFD vendors have offered such specialized (vertical) products for a long time. But they are usually for big niches, e.g., thermal management, ventilation, centrifugal pump.
As CFD software is used in more and more applications and industries, it is usually impractical to create a new product for each niche. It simply not cost-effective. CFD software vendors need the volume.
So, a good CFD software design is to provide easy-to-use building blocks for niche problems, and the software is a platform for all kinds of different problems and applications. Most CFD users are expecting tools that can solve “their” problems, not a “generic” problem.
Will we see some CFD software design change in this aspect in 2014? Probably.
Meshing actually is not a CFD problem. Meshing is not the purpose of CFD simulations, but most of CFD solvers cannot run without a mesh.
Generating a good mesh is not an easy job for real systems: it must represent the geometry, and it must capture the physics.
The meshing software usually understands (if it does) the topology of your domain (the geometry), but not your problem.
Either the CFD software or the user needs to know the physics to make sure the mesh is good enough to capture the major physics. Here comes the problem. Most CFD users are expecting good mesh automatically from meshing software (which expects the user to control and adjust because it does not understand physics). CFD solver is just expecting the good mesh from the input file.
The “normal” CFD workflow is somehow flawed for most users nowadays.
The traditional or normal workflow of “Geometry—>Meshing–>Physics setup–>Solving–>Post-processing”, is expecting some iterations. In other words, such a workflow requires trial-and-error. This wasn’t a problem in the old days. But this is the problem today.
Most CFD users are now expecting to get “correct” simulation in one-shot. This is not impossible. But it requires the involvement of the CFD solver during mesh generation. It also means a good mesh can only be generated after physics is defined. This is obvious because you need the mesh to capture physics.
Unfortunately, meshing software and solver still don’t talk to each other directly. They only communicate through the formatted mesh file/input file.
Will this change in 2014? Unlikely.
Open source CFD not only saves cost in software licensing, but also provides freedom you cannot enjoy from most commercial codes.
Academic users are the traditional beneficiary. They are the major contributors for open source CFD codes as well. The second statement may be not true for many other open source projects. For example, in last a few years, the primary code contributions for Linux Kernel are from big companies as reported here (you have to register to download the free PDF report, a sign of the invasion of commercialization of open source projects).
But in the commercial environment, the adoption rate of open source CFD is not as high as expected. Most companies will find out that the open source CFD solution may not save cost. I wrote a post last year explaining why open source CFD is not for everyone. Big companies can benefit more from the open source CFD than smaller ones.
One of the major obstacles of open source CFD adoption is the lack of high-level third-party support. There are a few companies sell support for open source CFD, but the scale is too small, and their technical muscle is not strong enough to convince potential users.
Who will wear the red hat for open source CFD? We probably will not have the answer in 2014.
So, in 2014, open source CFD software will probably still only make the rich richer.
After I wrote the cloud computing for CAE two years ago, a lot changed.
Now there are a lot of offers on the market. Some are reasonably priced with decent solvers. But the adoption is slow, and it will continue to be slow in 2014.
The problem is not in technology, but mainly in the mindset of users and software vendors.
For decision makers, the lack of the sense of “possession” may push back some potential buyers. The value of the software is in using, not possessing. But a lot of decision-makers don’t think this way. Of course, some IT nonsense on security or intellectual property associated with cloud computing is real nonsense. It just needs time to change the mindset.
For CFD software vendors, especially ones with an established installation base, they are reluctant to move to the cloud. Current licensing model probably is the safest way to keep the investors happy. Unless they are losing market share considerably, they will only be there to show “me too”. Some vendors who are actively promoting the cloud offering may get resistance from traditional distributors who will be partially redundant.
Of course, gradually, most users will move to the cloud and pay for what they use. This takes time, especially when changing the human mindset is involved.
Maybe, open source CFD software based cloud computing has better chance.
GPU is coming to CFD. More and more commercial code is or will support GPGPU to a certain extent. Nvidia even put a resource page on GPU for CFD.
But how much speedup GPU can offer for real cases is still a big question there. Most commercial CFD code still needs more tweaks to benefit from the GPU fully.
The requirement for high-end GPU card, uncertainty on speedup for real-world problems, and possibly more licensing fees will inevitably slow down the adoption.
Most CFD software vendors are now using webinar. This is good because webinar is convenient for two-way discussion, and it is effective for software demonstration and education.
But it is unfortunately used in a wrong way by those marketing geneses. Most vendors take the opportunity to brag, instead of branding, their software.
Eventually, such webinars will have the same fate as the whitepaper. Some years ago, whitepapers were usually respected and some have even been cited in academic publications frequently. But once marketing guys find there is a chance for bragging, whitepaper becomes toilet paper.
In 2014, you will see more and more webinars offered by CFD software vendors. Most of them are just wasting internet bandwidth.
What do you thin about CFD in 2014? Share your thoughts in the comment box below.
When open source CFD solution is talked, two familiar statements are often heard: Open source… Read More
Obviously, the doomsday didn't come. It proves again it is hard to predicate what will… Read More
We all know post-processing because it is part of CAE. All textbooks teach you how… Read More
Last year, FieldView jumped into Altair's bed, and now got a baby named AcuFieldView in the… Read More
Although there are quite a few mesh-free (mesh-less) FEA and CFD codes, meshing is still… Read More
If you happened to touch a real punch card in your university time, you probably… Read More
View Comments
Nice points. Another thing which we might see continue in 2014 is further consolidation of CFD space. Perhaps Ansys will try to acquire StarCD and create more monopoly. The way open source and cloud platform are growing, it might not be a bad idea for StarCD to sell.
Hi Shishir, thank you for the input. Actually, the rising of middle-tier (traditionally) CFD software may disrupt the status quo.
Nice year-end article with a good list of "food for thought and action" topics around the CFD focus. I think it's a good framework for a roundtable discussion. Maybe Ken Wong of Desktop Engineering or Jeff Waters, Group Organizer of "New Trends in CAE Simulation" would like to organize an on-line audio roundtable with some of us as participants.
Hi Dennis, thank you for the excellent suggestion.
Another nice article in the tradition of last year's predictions for 2013. And thank you for giving us the entire podium - 1st, 2nd, and 3rd place - for the problem of meshing!
Regarding GPU, we did an in-house study a couple of years ago. Although we were able to achieve fairly impressive speed-ups on certain operations (37x) the complexity of the coding changes would've made ongoing maintenance difficult (e.g. cross platform maintenance, accounting for lack of a GPU). Plus, GPU programming isn't simple especially from the memory management standpoint. Things have probably changed since then so we should probably re-investigate.
Thanks for writing this.
Hi John, thank you for sharing the precious info on GPU for meshing.
Really liked your "2013" post and from back then I anticipated the 2014 version.
Excellent points.
But I found quite interesting what you write about branding.
I infer from subsequent text that by branding you basically mean customer care / after sales support.
Do you believe CAE companies don't provide high-level support currently?
Plus you say, "Building trust with potential users is more and more important for CFD software vendors."
Do you also think that trust is not as high as it should be or you mean it has to go higher?
Secondly, you say "meshing software and solver still don’t talk with each other directly."
I thought adaptive meshing (provided now by most of vendors) was the solution to that.
Hi yiorgos, thank you for your comments.
For branding, it is more on "educate users". CFD software vendors need show potential users how to use their software to solver customer's problem. Bragging about the advantages or features will not build "trust".
For solution-based meshing adaption, it is a possible solution. But up to now, in commercial CFD codes, such mesh adaption still has very limited capabilities (compared to meshing software). Some of them can only perform simple refinement. Very a few can do mesh smoothing. Most of them ignore the inflation layers (therefore spoils your mesh). Some CFD vendors do not provide this at all. Even worse, a lot of CFD engineers never use this.
Well said about webinar Shengwei, Most cases it was used for Branding. We can belive only the top software that in market and their webniars...
I like your point about slow adoption of CFD (or CAE in general) in the Cloud. From my experience there are several obstacles that have contributed to the slowness. A couple that I'd like to mention here. The lack of cloud licensing models from the major commercial vendors is probably the main one. Only one vendor, that I know of, has a true cloud license. As a Cloud platform provider and integrator, we have to wrestle with all the "non-friendly" licenses models out there and it often just leaves the end-user frustrated. Another obstacle is lack of effective 3-D visualization from the Cloud. Visualization is critical to the work-flow and time can be saved by pre/post processing in the cloud where the data resides. Without good 3D (or even 2-D) capable visualization from the cloud, adoption will continue to be slow. We've overcome the 3-D visualization barrier but the licensing barrier is one that we can't control and it will take a lot of time to change.
Really interesting this post about CFD in 2014 being both an academic and an industrial user of CFD codes. I am curious to see the 2015 version.