XP vs Vista for FEA applications

[ThomasH]

Hi all

I suppose that you all have read tests compareing Windows Vista to Windows XP and the typical result is that Vista is slower. But is there any tests made for specifically FEM applications? I have seen tests with benchmark software but I can´t remember any tests with "real" problems.

Has anybody seen or done a test with 64 bit Vista vs 64 bit XP on the same machine, the same problem and the same software? Softwares like Ansys, ABAQUS, Nastran etc. The only variable shold be the OS.

It's just a curious question.

Regards

Thomas

 

[xerf]

ABAQUS has its own benchmark reports:

See:

http://www.simulia.com/support/v66/v66_performance.html

http://www.simulia.com/support/v67/v67_performance.html

Although, it is not exactly what you were asking for.

 

[cbrn]

Hi,
benchmarkings like that can end up being misleading.
In fact, the real big problem until some time ago was the "non-perfect" compatibility of FE programs with Vista. Now these issues are generally solved. So, provided you use the correct version of your preferred FE program, you shoudn't have any trouble.
As regards performance, consider that:
- Vista requires MUCH more power and system resources to run itself compared to XP, if you leave it with all the "gadget" features ON.
- It turns out being a sort of XP if you switch off these features (transparencies, animations, etc...)
- the speed at which the OS passes the low-level commands to the processor "should" not be different than XP.
- this is especially true when you speak about solution phase, because it doesn't involve GUI at all. Moreover, you can run a problem in batch mode, so that the influence of GUI disappears completely.

Regards

 

[ThomasH]

cbrn:

Your statements are exactly why I want to see benchmarks with OS as the only variable. And I mean that in a positive way.

Just as a comment: I have made tests with Vista in Classic mode (comprable to Win 2000/XP) and with Aero interface active. Not all the gadgets but some of them.

Impact on performance in terms of solution time? None.

Maybe I will reinstall my computer and test with XP but I'm not sure. But this is the reason why I would like to see hard data, not just some "common truth".

Regards

Thomas

 

[hondaknight1]

This is a very interesting thread, I have used vista64 on a clients machine and just found it so overhead heavy, we decided not to install it on any of our machines. But if its essentially the same as xp64 when you turn all the gadgets off then it may well be an option, especially as for new machines finding xp64 can be very difficult.

 

[cbrn]

Hi,
well, I understand it may be really boring, but if ANSYS is involved, perhaps the best way to make this benchmark is:
- "clean" a machine and install XP
- install Ansys and run a sufficiently "heavy" problem with the DIRECT SPARSE solver. Issue BCSOPTION,...,performance in order to expand diagnostics of the solution phase. Ansys will tell you repeatedly how many Mflops it is using, and how many "processor time" and "total time" (in seconds) it has used to achieve each phase. Before launching solution, redirect the output to a file. Keep this file somewhere
- "clean" the machine and install Vista
- install Ansys and run the exact same problem. Now compare the two diagnostic-output files.

I am personally very skeptic about "numerical benchmarks" which "simulate" the activity of a program via "standard test routines and low-level calls", because in many many cases they are far from the real-life.
As you have noted, I also personally think that a solution phase will NOT be slower in Vista than in XP.
Vista will be slower than XP in everything involving the GUI usage, at least until you don't deactivate the special features of Aero and you don't use an hardware / drivers specifically designed for Vista's compatibility.

Regards

 

[ThomasH]

cbrn:

Ansys is not involved, its NEiNastran. But you have the same basic idea as I have. I'm not sure about doing the clean/install cycle yet due to the work.

But since "everybody" claims that Vista is very slow I was curious if there was any hard data. I have a feeling that it might be a rumour that with repetition became a truth. I have seen benchmarks where Vista is slower than XP but they are of the "simulation of activity"-type. I was curious if anybody had tested the real thing. I have seen in other treads the "don't use Vista" statement. What is it based on?

Note that I'm not saying Vista isn't slower then XP. It might be but has anybody really tested?

Regards

Thomas

 

[cbrn]

Hi,
the rumour "don't use Vista" is still circulating "by inertia".
In fact, this statement was expressed by all those who tried to run Vista right when it was released, and hoped (I don't know on which basis) to run it seamlessly on the exact same machines which ran XP.
This is nonsense, because Microsoft had warned several times that Vista would certify drivers and hardware "designed for Vista". So most if not all the original problems with Vista are due to driver / hardware incompatibility. This is a thing to CAREFULLY check if you plan to install Vista on an ex-XP machine.
The same applies to FE programs: ANSYS has taken the time to rewrite some parts of their code to optimize for Vista, before the "Vista-compatibility" was released. RIGHT NOW, there is no reason why a correctly-configured machine with Ansys on Vista should run slowlier than an equivalent machine with XP. This is for analysis (solve) execute; of course the use of the GUI with Aero on can reveal a lot of drawbacks... This is not a joke: I've known several professionals reticent even to remove desktop themes or menu animations or things like that on their "calculation machine"... Look-and-feel may be important, BUT... work should be more, shouldn't it?

Regards