Computational Physics workstation! i7 rig (can be conviced otherwise)!

APPROXIMATE PURCHASE DATE: Sometime before Christmas BUDGET RANGE: €1000/$1500

SYSTEM USAGE FROM MOST TO LEAST IMPORTANT: Computational physics analysis, fluid dynamics simulations and media server

PARTS NOT REQUIRED: Keyboard, mouse, monitor and speaker.


PARTS PREFERENCES: Nvidia GPU is much preferred, as I will be running 64 bit linux only, and Nvidia generally have better Linux support. All parts obviously need Linux compatibility, but that usually isnt a problem these days. But as such, brand new hardware is often incompatible for a while so i suppose I should avoid new (< 2 or 3months old) hardware

OVERCLOCKING: A small bit, but stability necessary as could be running intensive code for days on end, crashing is not an option.

SLI OR CROSSFIRE: Not needed yet, but would like ability to upgrade to later perhaps.

Now I know this isnt exactly a European site, but I can track down parts this side, just need advice! As you can guess I'm just finishing a theoretical physics degree, and although I'm not set on a computational career, I do dabble in computational fluid dynamics. Although my first build in theory Ive been messing around with old computers for the past few years so consider myself not a novice, but a far cry from an expert. There is just a few areas I would like to get some advice from you kind folk.

I've pretty much got my mind set on the i7 920 x58, instead of the newer p55's as the future ability to upgrade is a bit too inviting. As i've said overclocking isnt a necessity although i will probably drive it up a little. Although if you feel otherwise do shout

Motherboards is one of the areas that i am not so sure on, I'd rather save a bit of money here, perhaps going for a mid-range board such as the ASUS P6T SE . Advice here is definitely needed!

Ram is another issue, In your opinions is it better to have 12gb low speed, around cl 9 , or 6gb high speed cl 7? price works out similar enough. Main brands corsair, OCZ, crucial etc ok?

Im thinking a 650W PSU, no ideas on brands, and this cooler, unless people have some issues
Hard drive will probably be 2 samsung 7200 1TB's, but thats not a large issue either.

My initial idea for the case was the near standard Antec 900 although It did occur to me that it may not be physically large

Finally in terms of GPU, my mind is not set. I picked this randomly to show price range and general area, Nvidia due to compatability, and something around
GTX 260 range

Well there you go guys, please give any advice criticism you can think of or forward me to any links you think would help. If you need any more info, do ask.

6 answers Last reply Best Answer
More about computational physics workstation conviced otherwise
  1. Best answer
    ^ You have pretty much selected the right kind of components required for your work...
    And even I wouldnt suggest the P55 over the X58 for the reasons that you have mentioned and also for the fact that the X58 CPUs offer Tri-Channel memory controllers(Extra bandwidth), which would be suitable for such tasks...

    And as for the overclocking, the i7 920 has enough headroom for 1GHz overclock...But as you want it to be stable, I would suggest you overclock it to say 3-3.2GHz, which must be easily doable at stock voltages...
    And do get an aftermarket CPU cooler, that would keep it cooler than the stock HSF...
    Scythe Katana 3 - Though it is not as good as the 120MM but can easily handle upto 3.6GHz without any issues...And comes at a very good price...

    Something better - Scythe Mugen 2

    And as for the mobo, that SE board is fine but it wont support SLI, if you want to...
    For that you can get the P6T...It is better power management and has SLI support too...

    And as for the RAM and graphics card, I would suggest this setup -

    Switch the graphics card to GTS 250 1GB(Would still be powerful enough for your computational needs)
    And with the money saved there, get 12GB of DDR3 1333MHz low CAS (Preferrably 7)...

    Or DDR3 1600MHz, CAS 8

    This combination would be better...And anyways there will be newer Nvidia cards available by Q1 2010, so later on you can upgrade to that...

    As for the case, this would be a very good option...
    CM HAF 932...Offers very good airflow, is a Full tower and at a very good price...

    Hope it helps...
  2. if your scripts can be multi-threaded, you should at least consider a multi socket Xeon workstation type configuration.. Gets you to between 8 and 16 cores pretty easily... Look at the Asus Z8PE or TYAN boards..

    Your budget, however, would be severely compromised.
  3. Some great tips there gkay09!
    Definitely going for the P6T and 12gb somewhat slower ram.

    Stil not decided between 1333Mhz CAS 7 and 1600 Mhz CAS 8. Very same price! Think one will offer a notable difference in performance, or even a small not so notable one?

    Splashing out less on a GPU sounds like a good way to make money, however I did fail to mention I would be working with visual molecular dynamics software quite regularly, sometimes up to 10k atoms and bonds on screen. From what I can see the GTS 250Gb should handle this but what you think? (upgrading later or SLI might fix that if id does indeed turn out to be a problem).

    Macmanes, A xeon was my first choice but expense as well as some personal history means i think ill be going i7.
  4. Are you running your own algorithms?
    If so, what are you running?
    If not, check with whoever makes the software you're running for advice on what to get.

    Whether to get 12 MB of slower or 6GB or faster RAM depends on how much you need. If you use over 6 GB, then the 12 GB will definitely be the way to go. No matter how fast the 6 GB is, swapping will destroy any performance advantage.

    Most computational fluid dynamics calculations will be some sort of finite difference type thing. Create a grid, all fluid flowing into each cell must equal the fluid flowing out. At least under steady state conditions. You have to conserve energy, etc... .

    For each run, you'll likely only use one thread since each iteration through the cells is dependant on the results from the last iteration, so you should have one core per analysis running in parallel. Plus a core for the OS. If you run one simulation at a time, you really only need 2 cores.

    Generally speaking, the 860 will outperform the 920 if you use 4 cores or less. So for strictly determining what's a better number cruncher, I would side with the 860. It uses a P55 motherboard, so you may lose some upgrade ability.
  5. Upgrade ability is something id like to have when it comes to cpu, Although i hope this computer will last me quite a few years, as intel seem to be promoting x58 for new processors I think ill have to stick with the 920. Hopefully once overclocked to 3-3.2Ghz the difference would be small.

    My area of current research is micro and nano fluid flows, so loads of numerical integrations and extremely large sparce matrices. The problem with only needing 2 cores is often a fluid dynamics calculation would be running alongside a analysis algorithm of my own, perhaps two and I would not like to compromise speed. Also it will be likely that often another user may be logged on remotely also running milder fluid dynamics algorithms.

    the extra speed gained from CAS 8 to CAS 7 must be small if not neglible? especially with 12gb ram?
  6. ^ Though I dont fully understand what you do, but I can certainly say this ...If your apps/ programs, are multi-threaded, then only you can utilize the full potential of the 920...But if core speed is more important, then I would suggest you overclock the CPU as the stock speeds are not pretty good...

    And as for the RAM, again if the tasks are more of bandwidth dependent rather than the quantity of RAM, I would suggest you get 6GB 1600MHz or higher RAM sets...
    And if you are going with 12GB, I would rather get 1333MHz CAS 7 RAM as it is easier to set 1333MHz when all the 6 slots on the mobo are populated...
Ask a new question

Read More

New Build Systems Product