Search
Sign in with
Sign up | Sign in

Number crunching

Last response: in CPUs
Share
Anonymous
CPUs Authority

Hello
Im building a computer for a fluid enginer who is running a large number of calculations/equations in an MS excel workbook. A typical equation would be somthing in the order of
H3-H4={1-D^2/D^2}V^2/2g+f*L/D*V^2/2g+V2/2g-V2/2g
This would be caluclated several thousands of times over

Firstly which CPU(s) would be the best for this type of work and is a dual CPU system a must?
Would a change in software increase performance?
Are there any other factors that would significantly improve performance i.e ram?

Thanks for your time
Lewis May

More about : number crunching

Register or log in to remove

RAM is probably unimportant, as the number of operations per memoryword is very high.

I think the software is the most important part. I don't know how good excel is at this, but I think the difference between the best and the worst software is greater than the difference between the fastest and the slowest processor.

Also. Make sure excel supports multiple threads in equation solving before buying a dual system. Actually, my system is dual, but I haven't got Excel installed. I'll come back to you with numbers later if I can somehow get a copy of msoffice.

/Markus
Related ressources
Ask the community

leme understand this, you got all THG people in there and you ask us?

heh, well being an AMD man i would have to go for the dual ATHLON MP 1800+

<font color=green>
*******
*K.I.S.S*
*(k)eep (I)t (S)imple (S)tupid*
*******
</font color=green>

Most cost effective would be a G4 733 or higher
Fastest would be MIPS 12Ka or a PMC in a O2 (being realistic)
Since its an engineer your building it for, go with the P4 478 and RDRAM. dont piss off the engineer trying to save a few bucks with AMD. you want to build him a bulletproof system not a time bomb.

Depends on whether your app is SSE2-optimized or not. If it is, the P4 has a chance of keeping up with an Athlon, but it will cost much more for a level of performance that's more-or-less the same as the Athlon. If not, your only sensible choices are an Athlon or a P3. Or, as FUGGER says, you could go with a completely different CPU architecture.

Careful though--if you go by FUGGER's advice, everything AMD manufactures is destined to go supernova in a week. =P

Kelledin

"/join #hackerz. See the Web. DoS interesting people."

I doubt it. I didn't see his name anywhere on the site (which doesn't mean a whole lot), and someone else that just registered posted with a red name in the Motherboards & Chipsets section. "Jaek"

<font color=green>I post so you don't have to!
9/11 - RIP</font color=green>

well when you are doing complex work like this it is not uncommon for a person to write their own program (especially with linux/unix systems) so it would be possible to compile the program with sse-2 optimizations. Personally I have heard that alpha's have the most killer fpu. But whatever you decide to buy stability is of the utmost importance so if you opt for an amd board be careful because their smp motherboards are new and unproven.

intel_inside your blurbing BS again, "approved"??
wtf are you talking about???
please dont respond to this, this is a retoricle question.
STFU!

<font color=green>
*******
*K.I.S.S*
*(k)eep (I)t (S)imple (S)tupid*
*******
</font color=green>

Ok, I do complex work. I use 3DSMAX- that takes shitloads of number crunching. I wouldn't even attempt writing 3dsmax- u have any idea how complex it would be? Although I know how it works (it's a constantly updating matricie), teh sheer amt. of programming would be crazy. Same thing for many other programs.

U got a problem?! Then dial 1800-328-7448!

If your friend is set on using MS Excel for iterative processing, then the FPU on the chip is not going to be the bottleneck. If your friend decides to code the equation in a higher-level language, he or she may need a highly pipelined FPU. The equation seems kind of odd. I don’t want to decode it, but there is a fair amount of redundancy and cancellation that needs to be done. If you have a fair amount of constants in your equation, you can look at vectorizing it. If this is the case, then streaming functions (SSE(2), 3DNow) are the way to go. Otherwise, a good assembly function will make a world of difference. Unless you’re talking about millions of operations, you could use a 386 to do it.

Schmide
Anonymous
CPUs Authority

Before you start worrying about the hardware you should try to opimize the code

H3-H4={1-D^2/D^2}V^2/2g+f*L/D*V^2/2g+V2/2g-V2/2g

H3-H4=({1-D^2/D^2}V^2+f*L/D*V^2+V2-V2)/2g

(i suppose the +v2-v2 is a typo)

V^2 and d^2 can be calculated once as you will most certainly use them more than once on the calculation

Give a real example of the code and I would be surprised if I could not reduce the time by 50%



BLL
Anonymous
CPUs Authority

Before you start worrying about optimizing *his* equations you should try and read the post *fully* :) 

Plus this guy is most likely well out of your league

“Plus this guy is most likely well out of your league”

If they were out of our league, they would know the answer to their question.

I’m ready to do some fxchg’n, its been a while but I’ll give it a shot.

Schmide
Anonymous
CPUs Authority

Talking about Leagues.

I certainly believe he is an outstanding engineer, but his profession is not programming and there is a very good chance he does not know how to fully use Excel and I dont think that it is a good idea for him to try to learn an other programming langue.

I am working for an international market leading company in fluid engineering since 1973 and my job is to clean up and fix the engineers programs (mostly Excel). I have spent 10 years writing programs for fluid engineering, programs my companys competitors does not even have yet. Actually there is a good chance I am playing in a league well above his.


BLL
Register or log in to remove
 
Ask the community
Read discussions in other CPUs categories
!