Thread Programming Hardware Configuration

Thread Programming Hardware Configuration

Bild des Benutzers p3tris

Hi, I've been programming using OpenMP and the Intel Fortran Compiler for some time now. I program under Linux environment (Ubuntu) and all the programs are tested on my laptop (2-core) or on a 4-core desktop.

The programs are engineering programs so they are computationally intensive and strictly mathematical (no web applications or graphics etc) I want to buy a "stronger" machine with more cores but keeping the shared memmory architecture. I need to be able to continue programming with OpenMP or in the future with TBB and continue programming under Linux. I received the following offer by a company:

  • 2 x processeurs Intel Xeon E5620 4 curs par proc. 2,4 GHz - Turbo up to 2,66 Ghz
    ou
    2 x processeurs Intel Xeon E5645 6 curs par proc. - 2,4 GHz - Turbo up to 2,66 Ghz
    ou
    2 x processeurs Intel Xeon X5675 6 curs par proc. 3,06 GHz - Turbo up to 3,46 Ghz

  • 5,86 GT/s. FSB 12 MB cache par processeur

  • Carte-mre Intel S5520HC Intel S5520 Server chipset

  • Mmoire de 24 GB UDIMM ECC 1333

  • Disques de 500 GB server class 64 MB cache 7200 rpm

  • Contrleur graphique ATI 16 MB intgr (basique)

  • Graveur DVD Sony 16 x interne

  • 2 x Onboard Gigabit ethernet connexions

  • Botier 4U Alimentation 850W + 4 x ventilateurs

Sorry for the french but I think you get the meaning. Is this formulation good for me? I mean the 2 XEON cpus see the same memmory? Will I be able to run my OpenMP parallelized programs under Linux like usual?

Any suggestions that will help to look better? Even any suggestion for a Shared Memmory architecture computer with more cores than 12? Thanks in advance, Petros

2 Beiträge / 0 neu
Letzter Beitrag
Nähere Informationen zur Compiler-Optimierung finden Sie in unserem Optimierungshinweis.
Bild des Benutzers Tim Prince
Best Reply

Yes, these Westmere dual CPU models have good support for ifort OpenMP or C++/TBB applications. The memory model is shared, with NUMA characteristics; that is, performance is affected by memory access local to the CPU vs. remote (memory local to the other CPU). Use of KMP_AFFINITY is important.
You may have noticed that the threading model of TBB isn't compatible with OpenMP and is supported only under C++.
The first significant further boost in mumber of cores and performance will be available with the Sandy Bridge dual CPU models to be released next spring.
The 850W power supply may be not be sufficient for more than the quoted graphics card, particularly with the X5675 CPUs. Mine has 1200W power supply.

Melden Sie sich an, um einen Kommentar zu hinterlassen.