open TELEMAC-MASCARET The mathematically superior suite of solvers

Hello,

I try to find an efficient way for me to make HPC on Télémac. So I have some questions:

Is it possible to install Telemac in a Windows server 2008 r2?

Can we use cards nvidia Kepler (CUDA) with Telemac ?
If no, can we adapt the code of Telemac to use these and how?
Or is that in the future Telemac code will be adapt to use these cards?

www.nvidia.com/object/cuda_home_new.html
www.nvidia.com/object/tesla-supercomputing-solutions.html

The future project of Intel : Intel Xeon Phi, could be used on Telemac without changing the Telemac code? Taking into account it use x86 instructions.


www.hpcwire.com/hpcwire/2012-08-28/intel....._a_little_bit.html
I know it is a lot of questions… I just hope somebody can help me on these topics
Thank for your help

Regards

Marine

Hi marine

I will say yes to all of your questions.
You have the sources file so you could do what you want.

Intall telemac on windows server 2008 is possible. This have been made 2 years ago by microsoft people for a test.

About CUDA, I don't know if it's already done but why not? the source code is written in fortran so a compiler should compile it in order to use GPU capacities.

same answer for intel project. The source code didn't use x86 instructions, this is the compiler which generate x86 programs but we already generate 64 bits versions... It's just a compiler configuration...

Hope this helps

Ok thank you for your answers
So acccording to you if the compiler works with CUDA or Intel projet there is no problem for Telemac. It could be running with the both.

Regards

Marine

Yes
You just have to take care about the various compiler and linker options.
If you reach to compile telemac you could easily check everything is OK by running the various validation cases.

Regards

Hello,

As a complement to Christophe's post, to run Telemac in GPU, need an important intrusive work on the code using OpenCL language for example (change in loops, conditions etc.). That's why, the use of Telemac for HPC with GPU is not so immediate.
The question is "why?", "is there any need for GPU for hydraulic applications like those of Telemac?"
In my humble opinion: the answer is no, at least for the moment and for the couple of upcoming years. For at least 2 reasons:
1- the parallelisation of the code with mpi is widely satisfactory. This is an opinion based ont he witness of a large number of users.
2- a standard size of Telemac models could reach the 1Million of node (at most). These models are efficiently handled by the parallel version and thanks to the fast resolution algorithm of Telemac, and this, even on standard personal machine equipped with a few number of procs.
3-GPU not matured yet (both langage and material) and platform dependant which is against the spirit of Telemac which aims to be a multiplatform, multi-OS open source code.

To be continued ...

All my best regards

Riadh

Hello Riadh,

Has this opinion changed over the years? On an outsider/user perspective, OpenCL/GPU computation has matured since and this could be a really interesting project to evluate.

Out of curiosoity I've been digging around asking for the state of art of this field for hydrodynamics and immediately Telemac was on my mind, hence this post.

Regards,

José D.

Hi Jose,

We tried to use GPU with the library MAGMA over the course of an internet ship.
This was a preliminary test but as of today no futher work is planned.

Yoann

Dear Yoann,

Thank you for the inside view regarding this topic!

Regards,

José D.

how is the performance?