Processors for modelling

Any thoughts on how different processors (specifically the Intel i5 vs i7) will affect modelling in FEFLOW?  I am in the market for new hardware and am not sure how to evaluate which processors are most cost effective.  Same for RAM.

I am most interested in cost effectiveness.  If the top end processor give 1% improvement in performance compared to the second tier processor, I am not that interested in paying for it.

Thanks!

Due to the short life times of processor generations and the high number of available types and clock speeds, we typically are not able to do formal comparisons between the different types. Therefore I am afraid we cannot provide you with a percental difference in speed between the i5 and i7 series. Furthermore, the difference might also depend on the problem type and solver used.

Thanks for the reply, Peter.  In your opinion, is there any way to give a generalization of which attributes have the biggest impact on modelling speeds - clock rate, FSB speed, CPU cache, something else?  What hardware do you use and how often do you update?

Clock rate influences run time nearly linearly on the same processor architecture. In general, we recommend to use mid-level processors (e.g., i7), we do not think that using technology on the upper end of the range is worth its price. We're currently using quad-core i7 processors on most desktop workstations, and computation servers with two hexacore processors. Update intervals are between two and four years, depending on the purpose of the machine (most often for servers, the least often for laptops that are only marginally used for actual simluations).

Hi,

Personally I did some test to see what is the effect of using multiple core for regional models (2 mo nodes). It seems that above 3 or 4 cores, the time-saving is very little. However if you plan to run several models at the same time as for PEST calibration, having multi-core is a really good improvements.

Cheers,
Robin

I use workstations with single-Xeon hexacore (12 threads) (W3680 or 3690) on server motherboards.  They are i7 with additional benefits: can use ECC memory, can access more memory larger cache, etc.  A bit more expensive than top of the line i7.  I run multi-day jobs, so ECC gives a bit of assurance that results are repeatable.  I agree with Robin, more than 4 core won't make feflow run faster, but if it is the only machine you have, many cores is a good thing.  You can run feflow while continuing work on other software and not have anything be unacceptably slow.  I figure these machines will last 3+ years.  Build your own for under $3K USD (off the shelf will cost 5 to 6K).  Dual Xeon workstations are very very expensive; both the cpu and motherboard needed more than double the cost...but the benefit is very small.  You can build two machines for less than the cost of one dual xeon.

Pete

[quote author=Robin Dufour link=topic=1070.msg2637#msg2637 date=1312493830]
Hi,

Personally I did some test to see what is the effect of using multiple core for regional models (2 mo nodes). It seems that above 3 or 4 cores, the time-saving is very little. However if you plan to run several models at the same time as for PEST calibration, having multi-core is a really good improvements.

Cheers,
Robin
[/quote]

Regarding 'to run several models at the same time as for PEST calibration', do you mean to run several FEFLOW simultaneously?
If so, wouldn't that cause license problem?
When I am using PEST-FEFLOW doing calibration, I cannot open FEFLOW because it will cause license problem to stop PEST.

Cheers
Ji

When running more than one instance of FEFLOW at the same time (even on the same computer), you will need a number of licenses available according to the number of instances you are running. To facilitate PEST runs, we will come up with a 'Short-Term Bulk License Rental' model very soon.