CPU selection for new workstation.

[DerwenArtos12]

System is for modifying extremely complext 3d models and editing simulation paramaters and runing very short simulations. Long simulations are uploaded to a mainframe for running. I'll typically upload anything more than 12 simulated hours, which takes about 45 minutes to run on my current workstation. Would like to be able to run 24 hour simulations inside an hour on the new workstation. Current worksation:

Dual Opty 285
Tyan motherboard(don't remember)
8GB ram per Proc(16GB total)
Four 500GB SATA drives using onboard raid 5
Dual Quattro's in SLI

Given the gudget I can start from scratch, from case and PSU on up. All the way through hard drives. For the purposes of being in CPU's and overclock just looking for CPU suggestions(though I'll take any suggestions you feel like giving me)

EDIT:Software running on it is completely in-house designed and has already been optimized for C2D for other engineers on this and other projects though not for 45nm as of yet, though I'm sure they wouldn't mind if it made me more productive.

Also, would like to build myself for the fun of it but, if there are any pre-builts out there that would make it pointless for me to actually do the build, I can refrain from insisting.

EDIT2: possibly more to the point, the 5450 harpertowns seem to be top dog on the block, what the heck is different between the three different versions besides price?

Thanks,

DA12

 

[myocardia]

The most expensive one, the BX80574E5450P, is an 80 watt. The other two versions are 120 watt. And I'm sure, like all of Intel's other server processors, the difference between the cheaper two is only whether it comes with a 1U or 2U heatsink.

edit: As for having 16 cores, you're most likely gonna have to wait until the 32nm Nehalems arrive.

 

[larciel]

Can I just watch while you build that machine and test it?

 

[DerwenArtos12]

Originally posted by: myocardia
The most expensive one, the BX80574E5450P, is an 80 watt. The other two versions are 120 watt. And I'm sure, like all of Intel's other server processors, the difference between the cheaper two is only whether it comes with a 1U or 2U heatsink.

edit: As for having 16 cores, you're most likely gonna have to wait until the 32nm Nehalems arrive.

I see that four quadcore opties completely rules out SLI or Crossfire. I'd be pretty content waiting another 2-3 months if that'll get me nehalems so long as I can still manage the build in 10K or under for 16 cores. I'm really tired of having to wait in the mainframe que(what I'm doing now)

 

[DerwenArtos12]

Originally posted by: larciel
Can I just watch while you build that machine and test it?

It's not going to have any publicly available operating system on it. It's going to have a custom linux distro on it that'll only run 2 programs, my simluator and the program to interface with the mainframe. I'll ask if I can take pictures of it since it's all publicly available hardware but, I'm 99.99999999999999999999999999999% sure they won't let me take my camera in as I'm not allowed to take my phone, keys or wallet in.

 

[myocardia]

Originally posted by: DerwenArtos12
I see that four quadcore opties completely rules out SLI or Crossfire. I'd be pretty content waiting another 2-3 months if that'll get me nehalems so long as I can still manage the build in 10K or under for 16 cores. I'm really tired of having to wait in the mainframe que(what I'm doing now)

Would you be content waiting another 18 months (minimum), until the 32nm Nehalems are released?

 

[Tullphan]

Originally posted by: DerwenArtos12
I'm 99.99999999999999999999999999999% sure they won't let me take my camera in as I'm not allowed to take my phone, keys or wallet in.

Dang...& I thought the prison I work at was particular about what we brought in! :laugh:

 

[DerwenArtos12]

Originally posted by: myocardia

Originally posted by: DerwenArtos12
I see that four quadcore opties completely rules out SLI or Crossfire. I'd be pretty content waiting another 2-3 months if that'll get me nehalems so long as I can still manage the build in 10K or under for 16 cores. I'm really tired of having to wait in the mainframe que(what I'm doing now)

Would you be content waiting another 18 months (minimum), until the 32nm Nehalems are released?

nope, lol. I have to have this built before May31st or it's not going to be able to get software loaded up till around christmas.

 

[DerwenArtos12]

Originally posted by: Tullphan

Originally posted by: DerwenArtos12
I'm 99.99999999999999999999999999999% sure they won't let me take my camera in as I'm not allowed to take my phone, keys or wallet in.

Dang...& I thought the prison I work at was particular about what we brought in! :laugh:

I have to have my badge and swipe it two times at seperate security checkpoints and then go up two floors and through another security checkpoint to use the bathroom. I see almost a dozen guards with MP5's before I get to my locker and about another dozen between my locker and my office. I got suspended without pay for three days because they did random inspections and I had forgotten to take my microSD card out of my phone. Even though it was two floors up in my locker, it's agianst company policy to have any type of removable storage device in the building. They could have terminated my contract so I guess I should be thankful I just got a long weekend.

 

[zsdersw]

You could go with this motherboard and 4 of the Xeon MP 7300-series CPUs, as seen here.. to get you to 16 cores now.

 

[DerwenArtos12]

Originally posted by: zsdersw
You could go with this motherboard and 4 of the Xeon MP 7300-series CPUs, as seen here.. to get you to 16 cores now.

No multi-gpu support. I modify highly coplex 3D models in real time and the guys that wrote this software have more or less told me multi-gpu is mandatory. Thanks though!

 

[Idontcare]

Originally posted by: DerwenArtos12

Originally posted by: zsdersw
You could go with this motherboard and 4 of the Xeon MP 7300-series CPUs, as seen here.. to get you to 16 cores now.

No multi-gpu support. I modify highly coplex 3D models in real time and the guys that wrote this software have more or less told me multi-gpu is mandatory. Thanks though!

So basically you need a Skulltrail.

It's dual-socket (8 cores in a box) but you can make up for the lack of 16 cores by clocking mutliplier unlocked QX9770 to higher core speeds. (8 cores at 4GHz ain't so bad, is it?)

And you get multi-GPU support.

 

[Rebel44]

Originally posted by: DerwenArtos12

Originally posted by: Tullphan

Originally posted by: DerwenArtos12
I'm 99.99999999999999999999999999999% sure they won't let me take my camera in as I'm not allowed to take my phone, keys or wallet in.

Dang...& I thought the prison I work at was particular about what we brought in! :laugh:

I have to have my badge and swipe it two times at seperate security checkpoints and then go up two floors and through another security checkpoint to use the bathroom. I see almost a dozen guards with MP5's before I get to my locker and about another dozen between my locker and my office. I got suspended without pay for three days because they did random inspections and I had forgotten to take my microSD card out of my phone. Even though it was two floors up in my locker, it's agianst company policy to have any type of removable storage device in the building. They could have terminated my contract so I guess I should be thankful I just got a long weekend.

If you want multi CPU multi GPU system you need Skulltrail.

OffTopic:
Where do you work ? NSA ?

 

[CTho9305]

Given what you claim to be doing, overclocking sounds like a really bad idea. Unless you have access to the manufacturer's complete testing equipment and reliability information, you're begging to produce a result where the CPU missed a decimal point or something. It also sounds like you may be in a situation where the best solution would be to test your application on CPUs from multiple vendors (AMD Opteron, Intel Xeon, IBM Power6, Sun's Niagara/Rock) and pick the one that's best for your specific needs. Keep in mind that the multi-socket scaling for Intel solutions is really horrible, so the best choice may not be obvious at all based on ordinary enthusiast site reviews.

 

[DerwenArtos12]

I will not be overclocking at all and given the preciseness of calculations needed, I would like to stick to workstation/server level hardware. Therefore I was asking about the Xeons.

CTho,
I know our mainfraim is IBM Power6 and they're considering upgrading depending on how some of IBM's new processors scale but, thats more than a year in the planning. Guys in other sections have already upgraded to various xeon workstations for their similar simulations and the software was completely re-worked when moving from the A64's like I have to the Core architecture so it's been optimized for these processors as best as our software engineers can. I know that they're slightly less hardcore machines(ie dual processor, 8 core, 8GB machines with single FireGL) can crunch notably faster than my existing workstation but, my modification and assembly times are still lower thanks to the extra graphics power I've got. From what I can tell anytime a simulation is believed to be more than 1/100th of a percent off, it has to be re-run and the system will automatically do so, up to 3 times and if the simulations are consistent I have to file a report and then spend the rest of my day figuring out what part of my design created the anomaly. Thats essentially what I do for 12 hours a day.

 

[Idontcare]

Originally posted by: DerwenArtos12
I will not be overclocking at all and given the preciseness of calculations needed, I would like to stick to workstation/server level hardware. Therefore I was asking about the Xeons.

CTho,
I know our mainfraim is IBM Power6 and they're considering upgrading depending on how some of IBM's new processors scale but, thats more than a year in the planning. Guys in other sections have already upgraded to various xeon workstations for their similar simulations and the software was completely re-worked when moving from the A64's like I have to the Core architecture so it's been optimized for these processors as best as our software engineers can. I know that they're slightly less hardcore machines(ie dual processor, 8 core, 8GB machines with single FireGL) can crunch notably faster than my existing workstation but, my modification and assembly times are still lower thanks to the extra graphics power I've got. From what I can tell anytime a simulation is believed to be more than 1/100th of a percent off, it has to be re-run and the system will automatically do so, up to 3 times and if the simulations are consistent I have to file a report and then spend the rest of my day figuring out what part of my design created the anomaly. Thats essentially what I do for 12 hours a day.

Still sounds like skulltrail is for you. Just don't overclock it.

No other mobo is going to give you dual-socket XEON with DDR2-800 FBDIMM and quad graphics PCIe slots.

And QX9770 (S771) are going to come native clockspeed still faster than the top speedbin XEON.

 

[DerwenArtos12]

Originally posted by: Idontcare

Originally posted by: DerwenArtos12
I will not be overclocking at all and given the preciseness of calculations needed, I would like to stick to workstation/server level hardware. Therefore I was asking about the Xeons.

CTho,
I know our mainfraim is IBM Power6 and they're considering upgrading depending on how some of IBM's new processors scale but, thats more than a year in the planning. Guys in other sections have already upgraded to various xeon workstations for their similar simulations and the software was completely re-worked when moving from the A64's like I have to the Core architecture so it's been optimized for these processors as best as our software engineers can. I know that they're slightly less hardcore machines(ie dual processor, 8 core, 8GB machines with single FireGL) can crunch notably faster than my existing workstation but, my modification and assembly times are still lower thanks to the extra graphics power I've got. From what I can tell anytime a simulation is believed to be more than 1/100th of a percent off, it has to be re-run and the system will automatically do so, up to 3 times and if the simulations are consistent I have to file a report and then spend the rest of my day figuring out what part of my design created the anomaly. Thats essentially what I do for 12 hours a day.

Still sounds like skulltrail is for you. Just don't overclock it.

No other mobo is going to give you dual-socket XEON with DDR2-800 FBDIMM and quad graphics PCIe slots.

And QX9770 (S771) are going to come native clockspeed still faster than the top speedbin XEON.

I don't need quad graphics, that would be fun and all but, I don't think the programs has been written to utilize quad graphics, I'll have to send and e-mail to our software engineers and find out.

Are there Skulltrail motherboards even available right now?

 

[zsdersw]

You could get one video card with two GPUs on it.. for the setup I mentioned earlier.

.. but Skulltrail is probably your best bet. If you *need* 16 cores now, though, you'd have to go the Xeon MP route if you want to use Intel CPUs.

You'd have to determine if more cores or more GPU horsepower would provide the bigger benefit.

 

[lopri]

What is the OS?

 

[Markfw]

I think This ASUS dual 771 motherboard with dual 5450 Harpertowns will get you almost 4x the performance you have right now.

 

[vailr]

Might consider a dual quad-core Penryn Xeon CPU: Mac Pro.
HP or Dell probably offer similar class dual-CPU Xeon workstations.
For the same money, the Mac Pro may be the better buy.
For video cards: you might save some serious money via bios flashing an "enthusiast" NVidia
or ATI GPU to become a "workstation" type GPU.
Similar cost savings on SATA Raid controller cards: Apple sells their SATA Raid controller for ~$1,000.
But: Newegg sells Areca SATA raid controllers for much less ($200 - $400).

 

[Idontcare]

Originally posted by: DerwenArtos12

Originally posted by: Idontcare

Originally posted by: DerwenArtos12
I will not be overclocking at all and given the preciseness of calculations needed, I would like to stick to workstation/server level hardware. Therefore I was asking about the Xeons.

CTho,
I know our mainfraim is IBM Power6 and they're considering upgrading depending on how some of IBM's new processors scale but, thats more than a year in the planning. Guys in other sections have already upgraded to various xeon workstations for their similar simulations and the software was completely re-worked when moving from the A64's like I have to the Core architecture so it's been optimized for these processors as best as our software engineers can. I know that they're slightly less hardcore machines(ie dual processor, 8 core, 8GB machines with single FireGL) can crunch notably faster than my existing workstation but, my modification and assembly times are still lower thanks to the extra graphics power I've got. From what I can tell anytime a simulation is believed to be more than 1/100th of a percent off, it has to be re-run and the system will automatically do so, up to 3 times and if the simulations are consistent I have to file a report and then spend the rest of my day figuring out what part of my design created the anomaly. Thats essentially what I do for 12 hours a day.

Still sounds like skulltrail is for you. Just don't overclock it.

No other mobo is going to give you dual-socket XEON with DDR2-800 FBDIMM and quad graphics PCIe slots.

And QX9770 (S771) are going to come native clockspeed still faster than the top speedbin XEON.

I don't need quad graphics, that would be fun and all but, I don't think the programs has been written to utilize quad graphics, I'll have to send and e-mail to our software engineers and find out.

Are there Skulltrail motherboards even available right now?

If you merely need a single PCIe 16x lane for a single graphics card then pick ANY dual-socket LGA771 workstation motherboard with a single PCIe 16x electrical slot and be done with it.

If you NEED high-end GPU (dual cards, either SLI or Xfire) then you want Skulltrail. Just because it has 4 PCIe slots doesn't mean you are obligated to fill them

Skulltrial (to my knowledge) is the only thing that is going to get you a dual socket motherboard which can also work with dual graphics cards.

 

[Idontcare]

Originally posted by: Markfw900
I think This ASUS dual 771 motherboard with dual 5450 Harpertowns will get you almost 4x the performance you have right now.

8x electrical on the fastest PCIe though, not going to help with the stated graphics requirements.

 

[GuitarDaddy]

Skulltrail Board $614

2 x QX9775 3.2ghz 12mb $3220

4x2gb ECC DDR2-800 $578

2x Quadro FX5500 $4600

Lian Li Case $250

Silverstone 1200w SLI PS $370

4 x Seagate sata 3.0 32mb 750gb $640

Optical Drive of your choice $35

Well that would exceed the budget at $10,307 +shipping, but something tells me $300 isn't going to be a deal breaker when considering a $10k budget
Actually you could scale the CPU's back to the harpertown 5450's and save $1200 and only give up 200mhz speed, if you are not going to overclock the unlocked multi and extra 200mhz isn't worth the premium for the QX9775's. You could also scale back the GPU's, you will understand your needs in that department better than us.


With the QX9775 I would have a very hard time leaving it a stock, a little multi bump to 9x400 3.6ghz wouldn't hurt anything


Sigh! I have absolutely no need for a machine like this, I know nothing about 3d
editing/rendering, but now I want to build one of these like really really bad

 

[Markfw]

Originally posted by: Idontcare

Originally posted by: Markfw900
I think This ASUS dual 771 motherboard with dual 5450 Harpertowns will get you almost 4x the performance you have right now.

8x electrical on the fastest PCIe though, not going to help with the stated graphics requirements.

I missed that....