Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.
Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
COMSOL on a dell T630 server, too slow
Posted 8 avr. 2015, 23:24 UTC−4 Version 4.4 3 Replies
Please login with a confirmed email address before reporting spam
hi,everybody.
I have just buy a new dell T630 server with 16 intel cores and 64GB RAM. COMSOL is running extremely slow on it and even slower than a old PC with only 4 cores and 4GB RAM. There must be something wrong with the configuration setting of COMSOL or DELL server.
Can anyone help me with the setting. Thanks very much!
I have just buy a new dell T630 server with 16 intel cores and 64GB RAM. COMSOL is running extremely slow on it and even slower than a old PC with only 4 cores and 4GB RAM. There must be something wrong with the configuration setting of COMSOL or DELL server.
Can anyone help me with the setting. Thanks very much!
3 Replies Last Post 17 avr. 2015, 09:57 UTC−4
Please login with a confirmed email address before reporting spam
Posted:
10 years ago
I hope someone can help you. I've been trying to get help with a similar problem. I spent ~$23K on a four-socket high-end computer with 512GB RAM, only to find that every problem I've tried on it runs much slower than on my 2-year-old single-processor cheap computer with 64GB RAM. COMSOL (for me 5.0, for you, I guess 4.4) seems completely incompatible with modern multi-processor high-end computers.
One of my 9.5M dof rf problems that ran in 30 minutes on my old computer using an indirect solver (GMRES) took more than 2 hours to run on my new computer using a direct solver (PARDISO). I've posted more details in another thread (high-end computer-guidance garbage), where there have been some responses, but none really helpful. I've been trying to find out who at COMSOL should be the point person on this, and hoping to hear of anyone who's not been disappointed using something more than a single-socket computer on very large rf problems.
And on another subject, I found 4.4 often gave completely invalid solutions to various rf problems that 4.3b and 5.0 solved perfectly.
David
One of my 9.5M dof rf problems that ran in 30 minutes on my old computer using an indirect solver (GMRES) took more than 2 hours to run on my new computer using a direct solver (PARDISO). I've posted more details in another thread (high-end computer-guidance garbage), where there have been some responses, but none really helpful. I've been trying to find out who at COMSOL should be the point person on this, and hoping to hear of anyone who's not been disappointed using something more than a single-socket computer on very large rf problems.
And on another subject, I found 4.4 often gave completely invalid solutions to various rf problems that 4.3b and 5.0 solved perfectly.
David
Please login with a confirmed email address before reporting spam
Posted:
10 years ago
Hello David,
What you are reporting here with respect to the direct and iterative solvers is expected behavior. The direct solvers will use more memory and as a consequence almost always be slower, than an iterative solver. Direct solvers do have their place for certain classes of RF problems, but when they can be used, the iterative solvers are superior in terms of memory and time. That being said, you will observe different runtimes for the same problem on different computer architectures.
You also mention here some other concerns that appear to be file-specific about invalid solutions on different versions. Such questions which are related to a particular file can be forwarded to COMSOL Technical Support if there is a confidentiality concern about the file contents. With each release we do verify that the models in the Model Library are consistent with earlier versions. The Model Library examples do include comparisons against analytic solutions to further ensure that the code is indeed solving the equations correctly. For us to comment we would need to see the exact files for which you are reporting this behavior, and will be happy to post our replies in this thread if you approve.
Best Regards,
Walter
What you are reporting here with respect to the direct and iterative solvers is expected behavior. The direct solvers will use more memory and as a consequence almost always be slower, than an iterative solver. Direct solvers do have their place for certain classes of RF problems, but when they can be used, the iterative solvers are superior in terms of memory and time. That being said, you will observe different runtimes for the same problem on different computer architectures.
You also mention here some other concerns that appear to be file-specific about invalid solutions on different versions. Such questions which are related to a particular file can be forwarded to COMSOL Technical Support if there is a confidentiality concern about the file contents. With each release we do verify that the models in the Model Library are consistent with earlier versions. The Model Library examples do include comparisons against analytic solutions to further ensure that the code is indeed solving the equations correctly. For us to comment we would need to see the exact files for which you are reporting this behavior, and will be happy to post our replies in this thread if you approve.
Best Regards,
Walter
Please login with a confirmed email address before reporting spam
Posted:
10 years ago
Dear Walter,
Thanks for the feedback. I probably shouldn't have included my comment about frustration with 4.4 on solving certain rf problems. That's history. My biggest current concern is precisely on topic for this thread, as we've discussed elsewhere. I'm really hoping someone out there will be able to recommend a computer that will actually run large COMSOL 5.0 or 5.1 problems several times faster than our current single-processor i7-3930K with 64GB RAM. We were expecting our new four-socket E5-4627v2 with 512GB RAM to be several times faster. Instead it is 30% to 50% slower (using 5.0 or 5.1). I guess the next step is to see if Linux is any better on our big machine. If not, and if no one can recommend a way to run large rf problems significantly faster than on a single-processor machine, a guess we'll evaluate CST. We really have to be able to take advantage of multiple processors, and so far we haven't seen any real evidence that recent versions of COMSOL can do that. Perhaps earlier versions could, though apparently 4.4 couldn't. If you can provide evidence that COMSOL 5.0 or 5.1 can take advantage of multiple processors, it would be most appreciated.
Regards,
David
Thanks for the feedback. I probably shouldn't have included my comment about frustration with 4.4 on solving certain rf problems. That's history. My biggest current concern is precisely on topic for this thread, as we've discussed elsewhere. I'm really hoping someone out there will be able to recommend a computer that will actually run large COMSOL 5.0 or 5.1 problems several times faster than our current single-processor i7-3930K with 64GB RAM. We were expecting our new four-socket E5-4627v2 with 512GB RAM to be several times faster. Instead it is 30% to 50% slower (using 5.0 or 5.1). I guess the next step is to see if Linux is any better on our big machine. If not, and if no one can recommend a way to run large rf problems significantly faster than on a single-processor machine, a guess we'll evaluate CST. We really have to be able to take advantage of multiple processors, and so far we haven't seen any real evidence that recent versions of COMSOL can do that. Perhaps earlier versions could, though apparently 4.4 couldn't. If you can provide evidence that COMSOL 5.0 or 5.1 can take advantage of multiple processors, it would be most appreciated.
Regards,
David