Thread Subject: Changing the MKL version in MATLAB

Hello:

I am using MATLAB R2008b. I think this version and the latest versions of MATLAB use the MKL library 10.0.3. My question is that I am using it on a Linux Machine. This machine has also installed the MKL version 10.1.1.

- Can I use 10.1.1 on MATLAB R2008b?

- If this is possible? How?

With many thanks in advance

JP

"Jes?s " <jpeinado.removethis@gmail.com> wrote in message <hahpjd$n98$1@fred.mathworks.com>...
> Hello:
>
> I am using MATLAB R2008b. I think this version and the latest versions of MATLAB use the MKL library 10.0.3. My question is that I am using it on a Linux Machine. This machine has also installed the MKL version 10.1.1.
>
> - Can I use 10.1.1 on MATLAB R2008b?
>
> - If this is possible? How?
>
> With many thanks in advance
>
> JP
I remember that with old versions, you could change the math libraries by just overwriting them in bin/glnxa64. It does not seem to be like that anymore.

Moreover: what has 10.1.1 to offer you above 10.0.3? Do you really need it, and/or is the advantage of increased speed more important than the stability of the Matlab package?

Sebastiaan

> Moreover: what has 10.1.1 to offer you above 10.0.3? Do you really need it, and/or is the advantage of increased speed more important than the stability of the Matlab package?
>
> Sebastiaan


Hello Sebastiaan:

Firstly, thank you very much for your answer.

Secondly. I want to compare the performance of an algorithm using MATLAB (with MKL) and the performance of an algorithm using MATLAB (with MKL) and also a hybrid code (using a NVIDA GPU and an Intel CPU). If I want to compare this, I think that the best way to do this, is using the same version (if possible) of MKL. The final idea is:

Alg 1 = MATLAB(MKL)

Alg 2 = MATLAB(MKL)+Hybrid algorithm (MKL+CUBLAS)

CUBLAS= The NVIDIA BLAS code.



But is I understand to you, changing the MKV version can have compatibility problems.

Thanks

JP

"Jes?s " <jpeinado.removethis@gmail.com> wrote in message <hak4fl$81n$1@fred.mathworks.com>...
> > Moreover: what has 10.1.1 to offer you above 10.0.3? Do you really need it, and/or is the advantage of increased speed more important than the stability of the Matlab package?
> >
> > Sebastiaan
>
>
> Hello Sebastiaan:
>
> Firstly, thank you very much for your answer.
>
> Secondly. I want to compare the performance of an algorithm using MATLAB (with MKL) and the performance of an algorithm using MATLAB (with MKL) and also a hybrid code (using a NVIDA GPU and an Intel CPU). If I want to compare this, I think that the best way to do this, is using the same version (if possible) of MKL. The final idea is:
>
> Alg 1 = MATLAB(MKL)
>
> Alg 2 = MATLAB(MKL)+Hybrid algorithm (MKL+CUBLAS)
>
> CUBLAS= The NVIDIA BLAS code.
>
>
>
> But is I understand to you, changing the MKV version can have compatibility problems.

I see. Unfortunately, Intel does not offer you to download older versions from their website, but I can offer you 10.0.5.025 or 10.0.2.018, that gives you at least a version closer to 10.0.3. (email me)

However, I do not think that there will be a (notable) difference for most of the BLAS functions, especially not the full-matrix blas. At least, I have never seen a measurable difference for BLAS level 3 operations between minor version updates.

The project sounds interesting. Any chance you will publish your results here? :)

Sebastiaan

"Sebastiaan " <s.breedveld@erasmusmc.REMOVE.BOO.BOO.nl> wrote in message <hak6l2$o26$1@fred.mathworks.com>...
> "Jes?s " <jpeinado.removethis@gmail.com> wrote in message <hak4fl$81n$1@fred.mathworks.com>...
> > > Moreover: what has 10.1.1 to offer you above 10.0.3? Do you really need it, and/or is the advantage of increased speed more important than the stability of the Matlab package?
> > >
> > > Sebastiaan
> >
> >
> > Hello Sebastiaan:
> >
> > Firstly, thank you very much for your answer.
> >
> > Secondly. I want to compare the performance of an algorithm using MATLAB (with MKL) and the performance of an algorithm using MATLAB (with MKL) and also a hybrid code (using a NVIDA GPU and an Intel CPU). If I want to compare this, I think that the best way to do this, is using the same version (if possible) of MKL. The final idea is:
> >
> > Alg 1 = MATLAB(MKL)
> >
> > Alg 2 = MATLAB(MKL)+Hybrid algorithm (MKL+CUBLAS)
> >
> > CUBLAS= The NVIDIA BLAS code.
> >
> >
> >
> > But is I understand to you, changing the MKV version can have compatibility problems.
>
> I see. Unfortunately, Intel does not offer you to download older versions from their website, but I can offer you 10.0.5.025 or 10.0.2.018, that gives you at least a version closer to 10.0.3. (email me)
>
> However, I do not think that there will be a (notable) difference for most of the BLAS functions, especially not the full-matrix blas. At least, I have never seen a measurable difference for BLAS level 3 operations between minor version updates.
>
> The project sounds interesting. Any chance you will publish your results here? :)
>
> Sebastiaan

Thank you very much. Well I did use the actual versions. About the results. No problem. I will publish the results when I have all the work finished. Thank you for your interest.

Jesus

"Jes?s " <jpeinado.removethis@gmail.com> wrote in message <hahpjd$n98$1@fred.mathworks.com>...
> Hello:
>
> I am using MATLAB R2008b. I think this version and the latest versions of MATLAB use the MKL library 10.0.3. My question is that I am using it on a Linux Machine. This machine has also installed the MKL version 10.1.1.
>
> - Can I use 10.1.1 on MATLAB R2008b?
>
> - If this is possible? How?
>
> With many thanks in advance
>
> JP

To come back to this problem: I forgot that you can actually change the BLAS library 'legally', by setting BLAS_VERSION:

On Linux Machines:
1. Download GotoBlas2 and install it
2. Bash terminal: export BLAS_VERSION=pathtoGotoBlas2/libgoto2.so
3. The same terminal: matlab

Victor's post reminded me of this:
http://www.mathworks.com/matlabcentral/newsreader/view_thread/261541

You can probably do something similar with the MKL library.

Sebastiaan

"Sebastiaan " <s.breedveld@erasmusmc.REMOVE.BOO.BOO.nl> wrote in message <hav6ej$iiq$1@fred.mathworks.com>...
> "Jes?s " <jpeinado.removethis@gmail.com> wrote in message <hahpjd$n98$1@fred.mathworks.com>...
> > Hello:
> >
> > I am using MATLAB R2008b. I think this version and the latest versions of MATLAB use the MKL library 10.0.3. My question is that I am using it on a Linux Machine. This machine has also installed the MKL version 10.1.1.
> >
> > - Can I use 10.1.1 on MATLAB R2008b?
> >
> > - If this is possible? How?
> >
> > With many thanks in advance
> >
> > JP
>
> To come back to this problem: I forgot that you can actually change the BLAS library 'legally', by setting BLAS_VERSION:
>
> On Linux Machines:
> 1. Download GotoBlas2 and install it
> 2. Bash terminal: export BLAS_VERSION=pathtoGotoBlas2/libgoto2.so
> 3. The same terminal: matlab
>
> Victor's post reminded me of this:
> http://www.mathworks.com/matlabcentral/newsreader/view_thread/261541
>
> You can probably do something similar with the MKL library.
>
> Sebastiaan

Thank you

Jesus