Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
10 juil. 2012, 10:08 UTC−4
I had similar cases.
For the same 3D transient heat transfer model, WINDOWS 7 took less time to finish the calculation than LINUX (SUSE).
I am planning to buy a new computer with WINDOWS 7 (64-bit).
I had similar cases.
For the same 3D transient heat transfer model, WINDOWS 7 took less time to finish the calculation than LINUX (SUSE).
I am planning to buy a new computer with WINDOWS 7 (64-bit).
Jim Freels
mechanical side of nuclear engineering, multiphysics analysis, COMSOL specialist
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
10 juil. 2012, 18:22 UTC−4
My experience with COMSOL on linux is quite the opposite. Perhaps it is your setup.
If you have a problem large enough to use 32 GB of RAM and then requires additional swap space, this could be the primary slowdown. While in RAM, whether it be linux or windows, COMSOL should run about the same speed as long as it is running 64-bit.
If your problem is so large that it requires swap space to run, and you are using a SSD for the swap space, perhaps the driver for windows has been optimized for the hardware, and the driver for linux is using a default driver which is not optimized and much slower. You might want to take time to investigate what driver the SSD is using.
My recommendation is to try not to ever use swap space. if your problem requires a hard drive to use swap space, then get a bigger computer, or use a cluster to take advantage of distributed parallel processing to remove the swap space. Swap space is too slow whether it is windows or linux.
My experience with COMSOL on linux is quite the opposite. Perhaps it is your setup.
If you have a problem large enough to use 32 GB of RAM and then requires additional swap space, this could be the primary slowdown. While in RAM, whether it be linux or windows, COMSOL should run about the same speed as long as it is running 64-bit.
If your problem is so large that it requires swap space to run, and you are using a SSD for the swap space, perhaps the driver for windows has been optimized for the hardware, and the driver for linux is using a default driver which is not optimized and much slower. You might want to take time to investigate what driver the SSD is using.
My recommendation is to try not to ever use swap space. if your problem requires a hard drive to use swap space, then get a bigger computer, or use a cluster to take advantage of distributed parallel processing to remove the swap space. Swap space is too slow whether it is windows or linux.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
11 juil. 2012, 06:21 UTC−4
thx!
the problem is that i'm working on my private pc because my institute doesn't have the money for a better pc. (we bought last week a new cluster with 256gb ram and 4 16core opteron but when we get it i'm already finished with my work)
i also can't reduce my mesh... (it's a 12 meter long systhem of 13 superconductor and a needed homogenety of 10^-4)
yesterday i tried a model which fits into ram (w7: 15min, linux: 19 min)
so the swap is also not the problem.
anyhow it's not such a big deal if i use the swap. (i need those results and if they need several days i've to deal with that)
i changed also some settings in linux for the ssd. (switched daemon, reduced swapiness and so on)
also the speedstepping/turbomode of the i7 works well under linux (my i7 gets automatically oc'ed to 3,2 ghz)
all in all:
cpu works well in linux (maybe another governor is needed?)
ssd is not the problem (only using the ram is still slower in linux)
could it be that the tempfile of comsol is the problem?
thx!
the problem is that i'm working on my private pc because my institute doesn't have the money for a better pc. (we bought last week a new cluster with 256gb ram and 4 16core opteron but when we get it i'm already finished with my work)
i also can't reduce my mesh... (it's a 12 meter long systhem of 13 superconductor and a needed homogenety of 10^-4)
yesterday i tried a model which fits into ram (w7: 15min, linux: 19 min)
so the swap is also not the problem.
anyhow it's not such a big deal if i use the swap. (i need those results and if they need several days i've to deal with that)
i changed also some settings in linux for the ssd. (switched daemon, reduced swapiness and so on)
also the speedstepping/turbomode of the i7 works well under linux (my i7 gets automatically oc'ed to 3,2 ghz)
all in all:
cpu works well in linux (maybe another governor is needed?)
ssd is not the problem (only using the ram is still slower in linux)
could it be that the tempfile of comsol is the problem?