Beginner evaluating SEMCADX, any one else with good info?
I'm evaluating SEMCADX for purchase, and I'm a beginner at numerical modeling, so there are likely incorrect impressions lists. Maybe I can post some of my experiences and impressions here to find agreement or be set straight. I am an RF engineer, a fair bit of experience with antenna modeling, written some code to do it. My goals are to do bio modeling for cancer research, which is why SEMCAD has been the focus. It was recommended by other researchers and in numerical analysis forums as was the FDTD method.
I do have to be careful not to post to explicit a details, but I need to know if many of these experiences are common to these packages? My suspicion is they are.
1st impressions
The SEMCAD crew came to town and was very knowledgeable about their product. They received a list of questions and demos which they complied with very well. With the lack of experience I have there were well thought out **** questions fired at them at the speed of light and they were very patient. Of course the demos were very fantastic which they always have amazing problems solved no matter what company. These problems included a lot of work on human body models, computing SAR, etc.
---------- Post added at 22:44 ---------- Previous post was at 21:31 ----------
Costs
To get the packages necessary for the goals required, the costs are astronomical (this seems to be common). Unfortunately about two times what was the desired spending. The pricing comes with many options because not all projects require all the capabilities. There is the initial software charge, there is a charge for support for the first year and ongoing years (seems like it should be included), costs for acceleration software, costs for additional packages to work with the bio models. The human bio models are themselves very cheap, part of the ITIS project Virtual Family and available strictly for research. SPEAG/SEMCAD were intimately involved in the model developments and SEMCAD has the capability to use these as pose-able CAD models.
---------- Post added at 22:48 ---------- Previous post was at 22:44 ----------
Licensing
There is a complicated licensing system (FLEXLM) which took support from SEMCAD to complete (my only experience is single seat installs of MATLAB, which I have no troubles). SEMCAD has licensing support for GPU cards (graphics PC cards which have 100 to 600 processors) through AXWARE, which allow much faster computation of numerical problems. Unfortunately the graphics cards that we have purchased are not supported by AXWARE, but an effort was made to include them in the release received for a demo. I think it has failed, though I think it seemed for work for a while. The supported cards seem to be dedicated numerical processing cards, rather than graphics cards and cost about S2000 per card. These can speed up computations of large problems by 50x.
The original solution to the GPU support problem was to change the environment variables, which was a well documented procedure. There were two “identical” machines that would be running the demo, though they didn’t behave the same way. I changed the environment variable on both machines and it would run the models using the acceleration on one machine and not the other. I didn’t confirm the patches on each machine or graphics processor software was identical between PC machines. Currently I don’t think it is running on either machine from the warnings window.
Multiple CPU support works with some of the packages and does not require an additional license. This alone improves computation speed substantially and may be enough to warrant not purchasing GPU support. The GPU support would be more effective at acceleration.
The GPU licensing continues to be a problem, but there could be an additional problem in the folks at SEMCAD are 8 hours ahead of me, so communicating over the phone isn't so easy. The problem could be I have two GPU cards and a monitor connected to each card. I should connect my monitors to one card and the other for GPU calculations. That is my next iteration.
Project
I am working on two models, a capacitive plate and a waveguide model. The waveguide model solves very quickly compared with the capacitive plate model. Both of these models have the same number of voxels, so I am not sure why one solves so much faster (my newby status shines through). I am quite happy with how easy it is to solve for S-parameters, though I have to review measured data vs SEMCAD calculations. Today I will also do a SAR and thermo simulation.