CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Hardware (http://www.cfd-online.com/Forums/hardware/)
-   -   Intel Sandy Bridge vs Westmere vs Nehalem (http://www.cfd-online.com/Forums/hardware/85221-intel-sandy-bridge-vs-westmere-vs-nehalem.html)

jakaranda February 20, 2011 08:49

Intel Sandy Bridge vs Westmere vs Nehalem
 
Hello,

I am looking to buy a high end laptop to do my meshing and pre/post processing on the go.
I am wondering which Intel architecture would give me the best performance.

It is not clear ot me which of the 3 intel chips (Sandy Bridge/Nehalem/Westmere gives the best performance for a laptop. I know that the Westmere is better then the Nehalem for CFD desktops applications (higher memory bandwidth and more cores for same price). But what about for laptops and what about hte sandy bridge chip?

Does anybody have any recommendations for laptops/manufacturers giving what they feel is currently the best high end bang for the buck?

Thank you

sail February 22, 2011 15:49

afaik, while cfd calculations can be efficently parallelyzed, meshing and postprocessing is usually done serially (at least for commercial programs) so you should not have a huge advantage in having lot of cores. i would instead give my preference to speed, cache, and architecture (plus, obviously tons of ram).

given this,by hearth i would go with sandybridge (higher clock speed, bus and newer architecture), but not enought data have been available, especially on mobile versions, about the single thread performance, probably because of the sata controller bug of the build 2.

as for a specific manufacturer I would reccomand dell. i have a precision m6500 and it is a joy to use. if you can wait a couple of months there should be the sandybridge version coming out soon...

abdul099 May 1, 2011 10:45

Nehalem and Westermere are pretty much the same but the production process and memory controller. Therefore I would use no Westermere, because the memory controller is no longer included to the cpu die, but only connected with a qpi. This results in bad latency when accessing the memory.
Sandy-Bridge should benefit most when the cpu has a lot of cores due to the difference in communication.

Opposed to sail, I think, more cores could help a lot. When meshing (a seriell process) takes long, it will be a big mesh and therefore the solver will need a lot of computing time as well and can benefit from more cores.


All times are GMT -4. The time now is 18:55.