[gmx-users] Gromacs using MKL with Intel 11.1 compilers
cousins at umit.maine.edu
Fri Sep 18 16:33:41 CEST 2009
On Fri, 18 Sep 2009, Vasilii Artyukhov wrote:
> Anybody have any real-world comparisons of using MKL vs. FFTW3?
> [ http://www.quantumespresso.org/user_guide/node16.html ]http://www.quantumespresso.org/user_guide/node16.html says:
> Axel Kohlmeyer suggests the following (April 2008): "(I've) found that Intel is now turning on multithreading without any warning and that is for example why their FFT seems faster than FFTW. For serial and OpenMP based runs this makes no difference
> (in fact the multi-threaded FFT helps), but if you run MPI locally, you actually lose performance. Also if you use the 'numactl' tool on linux to bind a job to a specific cpu core, MKL will still try to use all available cores (and slow down badly).
> The cleanest way of avoiding this mess is to either link with
> -lmkl_intel_lp64 -lmkl_sequential -lmkl_core (on 64-bit: x86_64, ia64)
> -lmkl_intel -lmkl_sequential -lmkl_core (on 32-bit, i.e. ia32 )
> or edit the libmkl_'platform'.a file (I'm using now a file libmkl10.a with:
> GROUP (libmkl_intel_lp64.a libmkl_sequential.a libmkl_core.a)
> It works like a charm".
> So, this might contribute to your problem. Please tell us if Axel's suggestion works for you!
Thanks for the suggestion. You read my mind! I was wondering about turning
threads off in MKL and here it is!
And here are the results: MKL without threads:
and MKL with threads:
As with FFTW3:
Even with the big improvement, since Mark says that only a fraction of the
tests use FFTW, the 5.3 seconds of difference between FFTW3 and MKL
without threads is a bigger difference than it appears.
I'll take Berk's advice and just use FFTW3. Thanks for setting me
straight. It's good to have confirmation that I don't need to spend any
more time on this! I just didn't want to use it blindly and then have
someone down the road ask "Why didn't you use MKL? It gives you twice the
performance!" or something like that.
More information about the gromacs.org_gmx-users