I'm currently evaluating CFD software for electronics cooling. I know that others have been down this path already and I am looking for feedback on your evaluations. Specifically, what are people's thoughts on Icepak, Flotherm, and Coolit.

I have been a Flotherm user for over seven years and have been pleased with it. I have reviewed the others on several occasions and have never seen a reason to switch. Flotherm's solver has always been rock solid and their software in general has been free of serious flaws. Their latest version is a comprehensive make over from their original product. It has what you would expect in a modern GUI, although I must admit from the perspective of a long time user it did take me a while to get use to the new front end. Old habits die hard, but once I got over the learning curve the CAD like front end is great. It is so much easier to make changes to existing models as well as create new geometry on the fly. The new users in my area learned the basics of the software in a couple of days and were able to produce usable results quickly. But beyond this, Flomerics has been in the electronics cooling buisnes now for 10 years. They clearly have the edge on industry experience. This adds a lot more value then what you would think. They have been involved in a grass roots effort to get component manufacturers to make available standard models of their components and have developed the ability to produce component level models quickly. Flomerics efforts in things like the Delphi project ( basic research on compact component level modeling) drove this. They are willing to spend some of their engineering $ in areas that do not always show up in the GUI but add value to thermal design community.

I have been an IcePak user for 1-½ years. We do packaging of UPS power supplies. 100watt to 2 Mega Watt. IcePak has given us reliable and accurate solutions, all which correlate to lab data. Models are generated quickly and simply. Solutions also come quickly. It helps us reduce iterations on lab tests and overall development time. We have 2 seats now and plan more in the future. My last look at Flowtherm indicated to us that it was not user friendly. I have not looked at Coolit.

Over the past 10 years I have had the privilege of using many CFD and FEA tools while working on electronics cooling design and analysis. Starting with Ansys I worked on complex transient spreading problems and PCB board designs. Over the years I found this tool harder and harder to use when I needed to look further at interactions with the overall system and solve complex fluid dynamic issues while understanding the effects on individual components. When the Flowtherm package first came out I quickly became a user and in fact purchased several seats for the design group I managed at the time. I must admit the front end was easy to use and the solution times were impressive. Unfortunatly, over time my group started realizing the limits of the mesher and solution methodology. As our problems became more complex we found that the solutions provided by Flowtherm were simply too general. To get the accuracy we required we needed to use complex unstructured mesh, something that was not available from Flowtherm. This is when we started to evaluate the Fluent software. Clearly, the Fluent package was a far superior product in it's ability to create complex mesh structures and provide accurate results at all levels. Unfortunatly, it was hard to use at the time. That has all now changed! Fluent's purchase of FDI, and the subsequent release of IcePak with the Fluent engine created a dedicated electronics cooling package with virtually unlimited mesh and problem solving capability. Now we have a tool that can expand with our problems and designs. I am now at Applied, an Aavid company, and I have direct access to all the IcePak and Fluent software for use in our consulting business. We use IcePak exclusively. It is easy to use, allows quick generation of detailed electronic components, cards, heatsinks, and systems. With the ability to input CAD details, materials, fan curves, and component models we can build complete system models in very little time. Each new release has added features that not only make it easier to use, but also allow more complex problems to be solved. With the backing of Fluent's software developers and a long history of CFD analysis expertise, I beleive it is more important to know what the tool is capable of and not what market is has been sold to.

Craig, working for an Aavid company (parent company of Fluent) did you have any choice but to use (or expound the virtues) of IcePak?

Mr(?). Pascarella, trial all 3 packages, push them all to their limit and compare them on the following:

useability stability accuracy support

I'd personally also look at the number of man hours of dedicated thermal analysis experience in the 3 companies.

(1).It is probably hard to evaluate CFD software without setting some reference parameters. (2). First of all, you can't ask the software directly to get the answers you want. So, it largely depends on the person who is using the software to answer the questions. (3). For the software, most people would like to know : a). The cost of the software. This is a must. b). The hardware required. Whether it is a Unix workstation, or a Pentium PC. c). Is the software self-contained, that is no additional software required. Like many CFD software, most of the time, one has to use a CAD program, a Mesh program, and a post-processing graphics program. d). What is the total cost of the software, and what is the total cost of the hardware. e). The time involved to create the model and set up the conditions for the solver. Most of the time, this represents 90% of the work. The processes involved in this pre-solver phase. f). The required experience and training for the engineer to successfully obtain the final result, one year, two years, five years, or ten years. g). The least important part is the accuracy of the results. This is because, if the result is not accurate enough for his applications, he is not going to use it. ( unless he has no choice. ) (4). I think, these are important factors to be considered when evaluating CFD software for industrial applications. But, if one is evaluating the accuracy of the results, then one has to design a common test case first. In this way, the programs under evaluation can be tested under the same conditions. Even in this case, the validity of the results is still limited to the scope of the test case used.

We regrettably severed our 5 year assoc. with FLOMERICS and switched from FLOTHERM to ICEPAK. This followed several hundred man-hours of evaluation.

I am not familiar with COOLIT; however, I have experience with Electroflo (TES; Troy, MI).

This thread could quickly turn out to be a "lets see how many champions Fluent can get to sing the praises of Icepak". This is not the most desirable method to convince someone of the benefits of a s/w package. The fact remains that Flomerics is still the market leader in this sector by a very wide margin and has continued to be so despite the best efforts of Icepak over the last couple of years. The benefits of Flotherm go way beyond the functional tickbox method of s/w evaluation often used by those who ignore such attributes as useability, integration, support dedication, software stability etc.

Mr. Pascarella (if you're still following this thread), there is much much more to effective electronic thermal analysis than a few fancy turbulence models and a supposedly useable fully automatic unstructured gridding system.

Robin Bornoff

Senior Consultant Engineer

Flomerics Ltd.

(1). The discussion has given me the opportunity to visit the site of Flotherm and Icepak. I think it is important to visit these site, even if one is not involved in the electronic cooling field. (2). About eight years ago, I had an opportunity to see how engineers at a research institute was doing in modeling the heat transfer problem at the chip and board level in Taiwan. The code handles the conduction and convection of the board with chips on it. The chip information was stored in a data base. (3). I think this is a very important field, especially as the level of integration of circuits in a chip or chip set start to increase quickly.

...indeed John. Package level modelling is a crucial market. It is often the case that the end customers of component manufacturers require representations of the packages to perform accurate computational thermal analysis. However, the component manufacturer does not normally wish to divulge the geometric construction of the package that is necessary to predict the junction (i.e. silicon) temperature (that which dictates thermal mode failure). Flomerics identified this problem many years ago and as a consequence was instrumental in coordinating both the DELPHI and SEED projects whose aim was to provide a methodology whereby boundary condition independent thermal resistor network representations of the actual packages could be supplied to the end customers. This satisfies the requirements of the end customer whilst allowing the component manufacturer to retain propietory information regarding the design and construction method of the package.

Flopack is Flomerics' implementation of the findings of the DELPHI and SEED projects. This dedicated www based product is indicative of the level of commitment of Flomerics to the electronic thermal analysis industry, a commitment that is simply not met by other general purpose commercial CFD s/w vendors who want to muscle in to make 'a quick buck'.

(1). I like your approach of specialized application. (2). Based on my experience, it is more productive to use specialized codes in routine design and analysis work. (3). But for non-standard problems, a general purpose code is normally used. In this case, the need is to obtain a solution rather than to increase the productivity. The overhead in the general purpose code is much higher than that in a specialized code. (4) Thank you very much for the information.

I did exactly the same thing you are doing now for Black and Decker 6 months ago. We had decided that we needed a CFD code for electronics cooling, so I set out investigating what was out there. Eventually I narrowed the field down to 3 packages; Flotherm, IcePak, and Coolit. I spoke to each company in order to have them come in and demonstrate their package. As part of that demo, we asked them each to solve a particular problem that we had been working on.

With Flotherm and IcePak, we were very impressed with the results we saw with the given problem. With both of those packages, results were within a couple of degrees of what testing had shown. We were also very impressed with the level of support each of them offered as well as the ease of use with the package.

The Coolit package on the other hand was another story. Fist of all, with Flomerics and Fluent they were able to give us a good solution based off the information we gave them. With DAAT Research (Coolit), I spent 3-4 hours on the phone explaining the problem. Then when the DAAT Research came in to do the demo, they were off in excess of 100oC with solution. It then took them another 2 weeks of working on the problem in order to come back with a solution that was closer to what we had measured. It was however, still much farther off than the results we had gotten from Flomerics and Fluent. As a result, we pretty much dropped Coolit from consideration at that point.

At the end of December last year, we sat down and listed the advantages and disadvantages of the Flotherm and IcePak codes based on our needs. What we determined is that for our applications, one package had no real advantage over the other. We did finally decide to go with the Flotherm package based off of the fact that we have one person who had used it at IBM several years earlier and was familiar with the package. Other than that, we could have tossed a coin and made a decision.

Now, I have been using the Flotherm package for six months and have been very happy with the results/support we have seen so far. We are still in the process of verifying simulation data to actual data to determine the accuracy/correlation of the simulations. Flomerics has also done an excellent job with the support through training, online help, phone help, and visiting our facility. Later this year we will be reviewing the Flotherm package to decide whether or not to renew the license or to try a different package. At this point in time, I don't see any reason for moving away from the Flotherm package.

(1). Without knowing exactly ( or vaguely) the specific problem you are interested in ( or working on), it looks like that one has to go through the same process over and over again to find a suitable code. (2). Would it be possible for you to identify the problem you are working on? What is the degree of difficulty? What was the need to use a CFD code? Was that benchmark testing results applicable to other users? (3). We definitely would like to hear your comments on these questions.

Essentially the problem I have been working on is an electronic power supply. The power supply is roughly 100mm x 75mm x 75mm. Inside this enclosure we are dissapating about 75W. There is a fan inside the enclosure, however due to the space constraints, it is butted directly up against some of the components severely affecting its performance. The 65W of the 75W is divided between 2 heat sinks, 1 dissapating about 40W the other around 25W. The remaining 10W is coming from other components in the enclosure.

The difficulty in solving this problem comes from the fact that the fan is very highly restricted as well as the need to control airflow in order to provide adequate cooling to each of the heat sinks. In each of the problems we have solved with the various CFD codes, we have been solving around 100K+ grid cells.

We recognized the need to use a CFD code because we needed a way to find what the airflow inside the enclosure was doing. Turbulence off the fan plays a large role in the cooling of the electronics. Extensive baffeling is also necessary to insure that airflow takes a path through our heat sinks. We also were very concerned with dead areas in the flow and areas of recirculation.

Whether or not the specific problem I have outlined is relevent to other uses is probably not important. We have been using the Flotherm code in ways that even Flomerics has been surprised about. What I think is important is the correlation of simulation results from each of these packages to actual test data. Before any of the 3 vendors we gave this problem to came in to do their demos, we had already done roughly 10 months of thermal testing. By the time they came in, we had a considerable amount of data for the thermal performance of the unit. When we gave them the problem, we asked them to solve it before some of the baffeling had been added to the enclosure. We then asked them to show us how one would add a baffel into the enclosure and to resolve the problem (if time allowed). In the Flotherm and IcePak case, they were able to show us results on the original problem that were very close to what we had seen in testing. They were both also able to quickly modify the problem and resolve. Again, they were able to show us results very close to what we had seen in testing. As a stated before, DAAT Research was not able to quickly and accurately provide the same results. This is what caused us to move away from the Coolit code. Also, because we are using these packages in ways that are different than what they were designed for, we felt that strong user support/training was a necesssity. Again, this pushed us towards the Flotherm and IcePak packages.

(1). I think this is an excellent example of how CFD should be conducted. (2). The user understands his own problem. (3). The user knows how to conduct the experiment and collect the data. (4). The user had the solution ahead of the time about the use of baffeling design. (5). The user took three approaches instead of just one. (6). The user allowed enough time ( 10 months of testing) to become familiar with his problem. (7). This is a perfect case that the user is well prepared and knows what he is doing. (8). The complex internal heat transfer problem with two heat sinks does require the right tool to get the good result. (9). I would say that 85% of the success comes from the user, and 15% of the success is related to the right code selected. Thank you very much for your information and time.