[SciPy-user] numpy/scipy on an Intel(R) Core(TM)2 CPU and CPU THROTTLING

Nils Wagner nwagner@iam.uni-stuttgart...
Mon Feb 19 01:33:43 CST 2007


David Cournapeau wrote:
> Nils Wagner wrote:
>   
>> David Cournapeau wrote:
>>     
>>> Nils Wagner wrote:
>>>   
>>>       
>>>> Hi all,
>>>>
>>>> I am going to install numpy/scipy (svn version)
>>>> on an Intel(R) Core(TM)2 CPU. And I would like to build 
>>>> everything from scratch
>>>> (including BLAS/LAPACK/ATLAS).
>>>>
>>>> Which compiler options/flags should I use for BLAS/LAPACK 
>>>> in that case ?
>>>>
>>>> Which fortran compiler g77/gfortran is
>>>> currently recommended to build numpy/scipy ?
>>>>
>>>> BTW, I am using openSUSE 10.2.
>>>>
>>>> Nils
>>>>   
>>>>     
>>>>         
>>> For Atlas, you should use the flags set by the ATLAS build system. If 
>>> you use recent ATLAS sources (3.7.* serie), there are some arch default 
>>> for your CPU, I think.
>>>
>>> cheers,
>>>
>>> David
>>> _______________________________________________
>>> SciPy-user mailing list
>>> SciPy-user@scipy.org
>>> http://projects.scipy.org/mailman/listinfo/scipy-user
>>>   
>>>       
>>  Hi again,
>>
>> I am confused about CPU THROTTLING.
>>
>> How can I switch off CPU throttling on SuSE Linux 10.2 ?
>>     
> I don't know anything about Suse, but if the command line cupfreq-set 
> works, you should try (as root generally) as written in the docs:
>
> cpufreq-selector -g performance
>
>   
On SuSE Linux you have

cpufreq-set -g performance

> cpufreq works by using so called governor (hence the -g), and the 
> performance one is more or less equivalent to disabling cpu throttling, 
> as I understand it (note that I don't know much about those things; but 
> if I do that on my laptop, it is working as expected).
>   
>> I found a command "cpufreq-set" but I don't know how to proceed.
>>
>> ATLAS uses gfortran by default but I have used g77 to build BLAS/LAPACK.
>>     
> ATLAS is written in C, so in my understanding is that the fortran 
> compiler is just used to build the Fortran interface to the C 
> blas/lapack built by ATLAS. I think the key point is to always keep the 
> same compiler (eg if you use gfortran, make sure you always use it for 
> all the libraries used by numpy/scipy like umfpack, etc...).
>
> One way to check which one is "better", would be to check other fortran 
> libraries you are using, I guess (eg, what is the Suse default: g77 or 
> gfortran ?).
>
>   
In the end I have used g77.
How do I build/install numpy/scipy with gfortran ?

If I just use

python setup.py build
python setup.py install

to build numpy/scipy g77 is called.


Cheers,

Nils

> cheers
>
> David
>
> _______________________________________________
> SciPy-user mailing list
> SciPy-user@scipy.org
> http://projects.scipy.org/mailman/listinfo/scipy-user
>   

 


More information about the SciPy-user mailing list