Obviously, the doomsday didn’t come. It proves again it is hard to predicate what will happen when you don’t have sufficient knowledge of how the system works now. Maybe it is just a bug in Mayan’s calendar (Google also forgot to put December in People App in Android 4.2 (aka Jelly Bean 4.2) last November).
Predicting the future of CFD is similar to weather forecasting, maybe just like Mayan’s Apocalypse.
But this does not stop us from expecting a new prediction, just like you check the weather forecast before travel.
As my first degree is meteorology related, please do not expect the prediction is more accurate than weather forecasting for the next 365 days.
#1. Automated meshing: a classical fairy tale continues
Fairy tales always have listeners, and of course, tellers as well. When you are young, you are listeners; when you are older, you may gradually become a teller. What has not been changed is the story.
Automated meshing is such a classic fairy tale in CFD industry.
Meshing is always one of the most challenging tasks for most real-world CFD problems (I am not talking about flow over an infinite plate). Automated mesh generation is the dream of a lot of CFD analysts.
So, here come many CFD software vendors bragging about their automated meshing in recent years. You will surely hear more in 2013.
As explained in Top 5 Misunderstandings on (good) Mesh, both geometry and physics must be considered when meshing. This implies that automated mesh is possible only if meshing is done after problem setup.
Unfortunately, in current (normal) CFD workflow, i.e., geometry–>meshing–>physics setup–>solving–>post-processing, meshing is done before physical problem setup. So, the meshing software has no information on physics when generating the mesh.
Moreover, so far in the market, there are no meshing tools (except solution based mesh adaption tools) has any intelligence on physics. So, even in some CFD software packages, especially, in CAD-embedded CFD packages, you may define the problem first, then click the solving. The automated meshing is still based on geometry only.
Until the meshing software gets CFD intelligence (this implies some changes in the conventional CFD workflow are necessary), fully automated meshing is always a fairy tale.
And this fairy tale will continue in 2013, with more tellers and listeners.
#2. CAD-embedded CFD: a placebo for the democratization of CFD
It is impossible to ignore the increasing demand (and expectations) for CFD simulations in almost all industries. The democratization of CFD gradually becomes a buzzword in this industry.
The proliferation of CFD simulations has a profound impact on the software development. One of the sign is that usability has become an essential factor in CFD software development as explained in “Ease of use, why is it important for CAE software?”
However, as pointed out by David Tatchell in 2009, simple-to-use CFD tools does not necessarily mean over simplified CFD (Don’t Mistake “Simple to Use CFD” for “Simple CFD”).
A lot of industry-specific CFD tools (some vendors name them as vertical products, as opposed to the general-purpose CFD) are doing impressive jobs by using jargon in these industries, automating some unnecessary manual work, deploying specific models, tuning solver parameters for particular sets of problems. This makes CFD an indispensable tool in product development in these industries.
Such simple-to-use CFD tools are part of the solution for democratization of CFD
Unfortunately, most CAD-embedded CFD packages only have “simple CFD”, which is often misused. A lot of people in this industry mistakenly link such tools with industry-specific tools.
Most CAD-embedded CFD tools are good for designers to visualize the flow or roughly estimate the performance of a not-so-complicated system. The problem is on neither the CAD software nor the CFD solver.
The challenge is to make a simple-to-use-CFD without sacrificing its capabilities for “general purpose” usage. In reality, most vendors cut all corners.
Making the simple-to-use CFD software requires intelligence on the problems to be solved. This is the reason why industry-specific CFD tools can succeed; while general-purpose CAD-embedded CFD tools are inferior in most cases.
There are no simple solutions for this yet, as the intelligence level of current CFD software is inadequate to give sensible recommendations or self-correction. There is a long journey ahead.
CAD-embedded CFD may tiptoe into your desktop already. But it is just a placebo for the democratization of CFD.
#3. CFD guerrillas: more comrades
One sign of the democratization of CFD, or de-specialization of CFD used by John Chawner and Mike Peery in the 8 Questions with Tecplot’s Mike Peery, is more and more non-analysts among CFD users.
Today, most CFD users are CFD guerrillas. They usually have an obvious target: “showing” CFD results in a short period (varying from a few days to a few weeks). CFD simulation usually is secondary in their job scope (“design” or “mechanical analysis” usually is the primary).
Software vendors have no choice but to work hard to provide some sorts of simple CFD solutions to cater to the demands from such users.
This trend will continue, and CFD guerrillas will have more comrades.
#4. Lattice Boltzmann equation based CFD: not a game changer in 2013
More than one year ago, I asked the question whether the lattice Boltzmann equation (LBE) based CFD could change the landscape of CFD in the short post of Navier-Stokes vs lattice Boltzmann: will it change the landscape of CFD?
Apparently, despite some clear advantages (as well as disadvantages) of LBE based CFD solvers, and the CAE marketing muscles of MSC Software, the market share for LBE based solvers is still too low to name it as an alternative to Navier-Stokes equations based solver.
It is no doubt, LBE based CFD software can hold its ground and inching into more industries dominated by N-S based CFD. However, we have to be a bit pessimistic about the speed.
LBE based solver will surely give N-S based solver vendors some pressure in the coming years, but the landscape of CFD will not change in 2013.
#5. Cloud computing for CFD or on-demand CFD: two turkeys do not make an eagle
If you follow the annual carnivals on supercomputing (e.g., SC series ) or HPC, cloud computing is an unavoidable term. It is also one of the fastest growing sectors in the IT industry: IaaS, SaaS, PaaS …
Cloud computing is not new for CFD. But major CFD software vendors are not interested in this area because of the business models (licensing) in this industry. An extended discussion on this was covered in the post of “CAE in the Cloud, Is It Just Hot Air?”
In the last two years, there are some positive advancements. One notable offering is Autodesk’s simulation 360 including the popular middle-range CFD solver, CFDesign (now known as Autodesk Simulation CFD). Altair has their HyperWorks on-demand (AcuSolve is included). Symscape has offered Caedium (OpenFOAM solver on Windows) on the cloud or bursting to cloud for some time already.
However, I didn’t find any financial reports or analysis on such offerings. It is very likely they are not able to punch their weight right now or in the near future.
Although cloud computing can be a deal breaker for the democratization of CFD, there are still quite a lot of obstacles ahead. Unconscious resistance to cloud computing in some organizations, conflicting interest of players in various sectors, and the business models of current CAE software company, should not be overlooked.
#6. GPU computing for CFD: Just a talking point
The superior floating point performance of GP-GPU makes it most popular hardware in a lot of modern HPC systems. NVidia’s latest K20 GPU has a peak performance of 4.58 teraflops (single precision), which are about two orders of magnitude of latest CPUs.
So, it is not strange most CFD vendors started to talk about GPU in recent years. Some even tried to support it. But unfortunately, the performance is far from ideal. Maybe OpenFOAM is the only solver thoroughly tested by the community with limited success.
Due to the memory capacity limitations in all GPUs, it is still difficult to fully utilize the power of all cores. In addition, multi-grid solver used by most finite volume method (and Navier-Stokes equations) based solvers does not scale well in a heterogeneous system (with CPU and GPU).
LBE based CFD solver apparently has better potentials in using GPU. But it seems the commercial LBE based software vendors failed to capture the momentum of GPU computing so far.
Any breakthrough in 2013? Very unlikely.
#7. Big data: no painkillers yet
Now it is not uncommon to perform a 10-million-element CFD simulation on a less-than-$2000-DIY PC. Apparently, you can easily get high fidelity results at an affordable cost.
The problem is the increasing volume of data generated. Similar to that in the IT industry, big data is a serious challenge in the CFD industry.
How to effectively manage and utilize big data from CFD simulations can make or break all simulation efforts. Storage is seldom an issue as the hard drive is quite affordable (less than 1 cent per GB for mechanical hard drives). The real problem is to manage and to explore big data.
Although most CAE vendors now have some sorts of simulation data management (SDM) system (some named it as simulation lifecycle management, SLM), the lack of exploring capabilities for CFD in most of these systems will make such systems less valuable.
As discussed in “From post-processing and visualization to simulation result exploration”, lack of simulation results exploration is the bottleneck of getting the full value of high fidelity simulations.
In 2013, we may witness some other offering similar to Tecplot Chorus, but big data for most CFD users are still the pain.
#8. Whitepaper=toilet paper
It is true that whitepaper is a marketing tool to promote certain technologies or methodologies or specific products. But today most whitepapers in CAE industry distort the reality to such an extent (to favour own products) that most readers would be misled with questionable or totally wrong arguments.
And to make it worse, some magazines and some so-called corporate blogs cite them directly without exercising any due diligence (sometimes, they have the same business interest and serve the same purpose).
The worst scenario is that many CFD guerrillas take such low quality (or no quality at all) whitepapers as the Bible.
In 2013, we will encounter more such toilet papers.
#9. CFD software price: far from commoditization
With the fast penetration of CFD simulations in almost all industries and the increasing number of CFD guerrillas, it is natural to speculate whether the CFD software market is commoditized.
Due to the complexity and diversity of CFD problems, and the current technology development in CFD, it is too early to speculate when the CFD software market will be commoditized.
As a side-effect of pre-commoditization, buyers do not have much negotiating power on pricing. You probably will not expect any price cut of CFD software.
Of course, almost all vendors will continue to enjoy the high profit margin.
#10. What’s your opinion?
Please leave your thoughts in the comments section below.
Chandra Bhushan Roy says
Good and honest post. Thanks
This post addressed big picture of CAE/CFD/hpc. Having worked in CFD and HPC interface at support level, I have to agree with misunderstanding most gorillas think as CFD understanding.
Now working in Marine Renewable Energy sector as grad student, openFOAM seems to be way for reliable and transparent solving of N-S equations. And who cares for cloud computing or any such word (market is mocking and cheating the researchers) , powerful machines and clusters are fine. CFD solvers are not like Angy-bird game. However big-data, would love to see myself stopped at such bottleneck.
Thank you for your valuable input.
Ries Bouwman says
Nice summary! Let’s see how much of this we can tackle in 2013.
Thank you for your comments.
Prophet for CFD!
Nice post, and very informative. I agree most of the 9 points, especially the ones on automated meshing, simple CFD and CFD guerrillas (I also cone of them!!).
Thank you for the nice blog, from which I learned a lot.
I am very glad to know this tiny blog is helpful for you.
Stan Posey says
Thanks for the thoughtful review, regarding #6 about GPUs:
1. ANSYS Fluent 15.0 in Q4 2013 will feature a GPU-based AMG solver with most benefit coming from coupled solver use
2. Commercial vendors FluiDyna (fluidyna.de) and Vratis (vratis.com) have GPU-based solvers for OpenFOAM today
3. FluiDyna also have a LBM product on GPUs called LBultra which can speedup a simulation by 20x (they report)
4. Other GPU developments exist and continue to improve such as AcuSolve from Altair, Moldflow from Autodesk, etc.
5. Nearly every commercial CFD vendor has either a GPU product evaluation or research project ongoing — please ask
Thank you for your detailed info on the status of GPU for CFD solvers.
Regarding the LBE: The LBE based CFD is not “a changer” in 2013 for the same reasons it was not in 2012. There are very few people dealing with LB, comparing with NS, as there are very few people dealing with Monte Carlo compared to NS. It is not because the LB method is not prepared, but the users are not prepared, and actually probably won’t be. You may need a generation change, to hopefully be “a changer”, so death will eventually make LBE based CFD ”a changer” 🙂
2 days back my boss asked about GPU instead CPU. came to know from your post that lattice Boltzman equation based software can make use of this. Need to know Fire dynamics simulator can make use of it.
If you are talking about FDS, GPU is not used, as I know.
Just a comment: There is an interest work on LB on http://dx.doi.org/10.1103/PhysRevE.88.013314, where CPU is used and not GPU.
Just a question:
You wrote: “As a side-effect of pre-commoditization, buyers do not have much negotiating power on pricing. You probably will not expect any price cut of CFD software. Of course, almost all vendors will continue to enjoy high profit margin”.
I’m little bit confused. The prices of commercial CFD has decreased in the last years. From a one-year license for about 25 000 USD to a perpetual license for the same 25 000 USD.
Some how, their margin profits has decreased a lot, Or?. If the “trend” continues they will be cheaper. Therefore, I don’t know from where you got the “vendors will continue to enjoy high profit margin”.
Please, I will be interested to know how you reason about your aforementioned statement.
Swapnil Gaul says