Work Machine/Beast

[CLite]

Hi all,

I'm building a machine to do large scale engineering solutions for Finite Element Analysis (FEA) for a program called Abaqus. Basically it's going to be crunching enormous matrixes using iterative methods. My company essentially told me to be reasonable with the budget, lets say $1500, I'm mainly buying the parts from newegg for ease of dealing with warranties/etc. I convinced my company that running these things on my work laptop ruins my productivity since it locks my machine and I can't do large scale models. Our Abaqus license only allows us to use 2 processors.

The goal:
A 2 Core beast that doesn't mind having 8 gigs of ram being maxxed. Say overclocked to 4Ghz + ?

The guts:
Mobo: Gigabyte GA-X48-DQ6
Cpu: E8500
Gpu: EVGA 8800GT
RAM: 8 GB G.Skill DDR2-800
HD: WD raptor (OS/Abaqus) + some WD 7200 drive.
Psu: Corsair 520HX

The peripherals:
P182 Case
Swiftech H20-120 watercooling

Above items total: ~$1200
Left over Budget: ~$300 (Don't feel obligated to spend this)
*edit* yeah... forgot to include HD's, remaining budget is more like $0


Just looking for basic thoughts like: that case/cooling combo is the stupidest thing ever, or recommendations on better equip. Really love these forums and hope I organized this post well enough for you all.

 

[DSF]

Why the 8800GT? Do you really need a lot of 3d graphics power?

I would skip the X48 motherboard. It doesn't do much for you, and you could save a fair amount without sacrificing performance by going with a solid no-frills P35 board. Same with the E8500. For $100 less you could have an E8400 which will overclock to just about the same point.

Which brings me to the RAM: If you're planning to overclock the chip as far as it will go, you're going to want DDR2-1000, not DDR2-800.

I also wouldn't bother with the Raptor, as the speed advantage just isn't there like it used to be. Get the Western Digital Caviar 640GB.

 

[Goldfish4209]

I agree with DSF. A P35 will be perfectly fine; they overclock pretty well. Also, you could also shave off a few bucks by going with a corsair 450VX. I'm running a similar rig on a 450VX with no problems. (6750 to 3.2Ghz, 4Gb RAM, stock 3870)

 

[ther00kie16]

You should look into whether or not that program utilizes quad cores. If it does, get a q6600.

 

[Blain]

If you're building for a business, stick with a stock speed quad core.

 

[0roo0roo]

don't overclock.
as said, its stupid if its business solution where stability trumps any slight increase in speed.

 

[CLite]

Thanks for the input so far guys,

1) Regarding the 8800GT, this program has a 3D modeling aspect that gets very sluggish if I create large models. The 8800 GT for $180 imho is a bargain deal compared to topline cards of the past.

2) I agree with saving money by going the E8400, thanks for that suggestion. The program can support arrays of 30+ processors (yes models can get that crazy) however our business only has enough "tokens" for 2 cores. I won't bother with the quad because of this.

4) I agree I don't want to compromise stability, I thought running prime tests etc. pretty much assured it was stable however. My goal was to tweak to 4Ghz with a top of the line intel core 2 duo, however if you think it won't be stable I certaintly won't do it. The reason I am trying to tweak this is because any time saved solving the larger models is beneficial to me.

This program costs us about 70,000 dollars a year to license, spending a few extra hundred to create a machine that utilizes it effectively and with stability is my goal. The program would use dozens of GB's happily and max out any processor combination out there for days if a model was complicated enough. My computational time is directly tied to how much active ram it can handle and the speed of the processor which is why I'm trying to tweak this as if it's a gaming rig. However as pointed out by the previous posters stability is king here.

 

[Goldfish4209]

"minor" overclocking probably won't cost you any stability. You could probably leave everything at default or auto, push the FSB to 400, and pretend it's a Xeon X5272 (in the case of the 8500).

 

[DSF]

So you need a separate license or token or whatever you want to call it for each core, not each CPU or each machine?

 

[CLite]

DSF:
As a byproduct of buying more seats for your company you slowly unlock more CPU cores that can be used for each "seat". This is kind of an oversimplification but the full explanation is pretty convulted. Basically we have 2 "seats" so we can have two people using it each using 2 cores. Big companies like car manufacturers who run crash models might be using a dozen linked cores because they have so many "seats".

Goldfish:
Would you recommend DDR2-800 or 1000 for "simple" FSB tweaking?

 

[Blain]

Running Prime95 is fine for playing around where data is expendable.
But again... For a business where data integrity is critical, Stick with a stock speed dual or quad core processor.

Scenarios...
1. You OC a build and get "free speed" from a lower cost processor...
A couple of people think you're a hero for saving the company a few bucks.
2. You OC a build and after a while something goes wrong and data gets corrupted along the way...
Everyone is steamed at you for costing the company money in down time, or loss of critical data.

Scenario 1. isn't worth the risk of scenario 2. :roll:

>> Stay with default speed on the CPU <<