Building a $1600 system for CUDA number crunching


I have a budget of about 1600$ before tax. I am building this system mostly for crunching number/ engineering software/ and some gaming. I am struggling to decide between i7 860 and i7 930. CUDA performance is quite important. I would also be doing some virtualization but nothing too hardcore. Please give me some advice. Thank you so much!


BUDGET RANGE: $1600 After Rebates

SYSTEM USAGE FROM MOST TO LEAST IMPORTANT: number crunching, gaming, engineering apps

PARTS NOT REQUIRED: keyboard, mouse, monitor, speakers, OS



All prices are in CAD.

LGA1156 build:
ASUS P7P55D-E Pro LGA 1156 Intel P55 SATA 6Gb/s USB 3.0 ATX Intel Motherboard $196.95
Intel Core i7-860 Lynnfield 2.80GHz 8MB L3 Cache LGA 1156 95W Quad-Core Processor - BX80605I7860 $314.95
2xPatriot G Series Sector 5 Edition 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10666) Desktop Memory Model PGV34G1333ELK 2x$113.95

ASUS P6X58D-E LGA 1366 Intel X58 SATA 6Gb/s USB 3.0 ATX Intel Motherboard $244.95
Intel Core i7-930 2.8GHz 8MB L3 Cache LGA 1366 Quad-Core Desktop Processor $308.95
G.SKILL PI Series 6GB (3 x 2GB) 240-Pin DDR3 SDRAM DDR3 1600 (PC3 12800) Desktop Memory Model F3-12800CL7T-6GBPI $191.99 + $9.73 Shipping from

I am getting the GTX470 card for some cuda power.
EVGA 012-P3-1470-TR GeForce GTX 470 (Fermi) 1280MB 320-bit GDDR5 PCI Express 2.0 x16 HDCP Ready SLI Support Video Card - 012-P3-1470-TR $382.95

Intel X25-V SSDSA2MP040G2R5 2.5" 40GB SATA II MLC Internal Solid State Drive SSD $129.95
Western Digital Caviar Black WD1002FAEX 1TB 7200 RPM 64MB Cache SATA 6.0Gb/s 3.5" Internal Hard Drive $109.9

Case and PS:
COOLER MASTER HAF 932 RC-932-KKN1-GP Black Computer Case With Side Panel Window $138.95
OCZ 700W ATX12V V2.2 / EPS12V SLI Certified CrossFire Certified Modular Active PFC Power Supply OCZ700MXSP $96.95

Any comments would be welcome!

1) LG1156 vs. LG1366
2) Is it better to save some money and skip sata6G and usb3?
3) should I wait for solid state drive to drop in price?
4) Is a 140$ case an overkill for this build? (but then, I'm in Canada, so a 80$ case becomes 130$ here)
5) Any advice would be welcome

Mainly CUDA number crunching for machine learning algorithms (datasets with millions of images, learning millions of parameters). So most of the crunching would in fact be done at the GPU.
MATLAB. not so hardcore here.
Virtualization: Most likely, I will install windows and virtualize ubuntu... or setup a dual boot since I need both OS

So, Lynnfield (i7 860) has PCIe directly on CPU whereas Bloomfield (i7 930) connect through X58 chipset. What difference does this make? Would Lynnfield be better for single GPU?

Thanks for the advice so far!
4 answers Last reply
More about building 1600 system cuda number crunching
  1. You want to go with the i7-930 build. It's a much better number cruncher. That said, I wouldn't get the OCZ RAM. OCZ's sticks have compatibility issues, especially with Intel. Grab some good G.Skill Pi Series sticks.

    I would get USB 3/SATA III support on the board, but not the devices. Right now, they're not a lot faster. Pick up a Seagate 7200.12 or Samsung Spinpoint F3 to save a few buicks.

    I would definitely wait on the SSD. They don't do a whole lot for actual performance (though they seem really fast), and are extremely expensive.

    A $140 case is always overkill. The HAF 922 is better at a much lower price.

    I'd also look at a better PSU. Check out some 750W units from Corsair, Antec, SeaSonic and Silverstone. OCZ is just decent quality. You really want to have a PSU with excellent quality.
  2. GO LGA 1366

    Get the fast ports ... this is a big investment ... do all possible to ensure longevity.

    Get the INTEL X25M/G2 80GB NOW ... as a boot/apps drive. Load OS ... be done.
    Wait until at least September, to purchase/add any others ... even a year.

    I do not understand question #4, at all. There are provisionas to edit your OP.

    That graphics card requires a nuclear power plant to run it ... check your current and future power requirements very carefully. If you plan on a 2nd GPU (of the same ilk), ... you gonna need a whoppin' big PSU.

    If you want to avoid overkill, you need to list any specific tools (or tool types) that you will be using the system for ... We can go and check your system requirements and project how much "compute power" you can make use of, in the next 5 years.
    ... For you to simply say, to me ... "Number Crunching" or "Engineering Type Stuff" just doesn't cut it, with me.

    Fact is ... We have many computers in the Chemical Engineering Dept., here, at the university ... except for the folks doing molecular modelling (will never get enough power), most of our legacy labs are using older AMD (like really old) dual cores and running XP ... Things change really slow, here.

    Across the street, in the Astronomy building ... They process large datasets ... and focus on fast storage and a snappy proc ... Caddy-corner, from us is the Mechanical Eng. Dept. and School of Architechture ... They do lots of CAD design.

    Not to talk to you like a baby ... but just to point out that, if YOU talk to ME like a baby, then I really am not going to be able to give the correct answers (recommend an appropriate build)...

    ... Regardless ... I would steer you toward the 1366/X58/Core-i7/930 with 1600c7 fast RAM and a decent cooler.

    If the top performance AMD platform is not enough power (yet to be determined) then skip completely over the 1156/i7-860 and go directly to the X58/930 option.

    Name your tools and give us an idea of how much and how often you are gonna pull what loads.

    Basically, you have told us you want a pickup truck for "construction type work" and for landscaping ...

    From this info ... we can't say if a crew cab is needed ... are you hauling leaves? ... or, logs? .... are you mowing around your townhouse, once a month?? ... or do you do a major golf-course, every day?

    ... OK ... enough baby-talk.
  3. thanks for the advice. see update in OP.
  4. So ... you think your computer might be able to "LEARN" how to go off and crunch things even when you are not sitting in front of it ? ... Sort of like large media renders? ... Sure, then 1366/X58/930(OC)/1066c7x12GB ... (unless you are a *struggling* student).

    You can shave costs by going with only 6GB of RAM ... 12GB prolly wouldn't hurt.

    What about the size of any local datasets? ... I just want to be sure that the drives don't choke you.

    Display/CUDA subsystem ... How many mons? ... Rez?

    You still have not named any specific tools or toolsets. Very hard to ensure the correct compute balance, without knowing which features and protocols and language extensions are to be employed.

    = Al =
Ask a new question

Read More

New Build Systems