• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

256-bit 2gb or 320-bit 1gb and CUDA core question?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Hookem

Member
Joined
Jul 3, 2001
Location
Houston Texas
Need some help choosing a video card. Build consist of a core i7 3770k processor (thank you MicroCenter) with the AsRock Z77 Extreme6 paired with 32gig's G.Skill memory.

http://www.newegg.com/Product/Product.aspx?Item=N82E16820231569

Box will be used to render video for live shows in HD, general office use and 10 to 12 hours gaming (wish I had that guys job).

But my main concern for the build is rendering video not FPS playing games.

Budget is 250 to 300 tops. Comparing two EVGA cards. One has 2GB 256-bit GDDR5 with 385 CUDA cores. The other card has 1280MB 320-bit GDDR5 and 448 CUDA cores.

Here are the two I am looking at.

http://www.newegg.com/Product/Product.aspx?Item=N82E16814130738

http://www.newegg.com/Product/Product.aspx?Item=N82E16814130683

Which would be better for rendering. More CUDA core with faster bus or More memory with slower bus and less CUDA Cores?

Also there is the Open GL 4.1 or 4.2 differences.

Any help would be greatly appreciated.
 
Last edited:
Without trying to pretend that I'm some kind of video card expert (audience laugh track here) I do know that the more bits available for throughput the better. It was described to me as a simple analogy:

Would you rather commute on a 256-lane highway or a 320-lane highway? More commuters per second on the 320, speeds being equal. And being in Houston I know you can relate.
 
Last edited:
Without trying to pretend that I'm some kind of video card expert (audience laugh track here) I do know that the more bits available for throughput the better. It was described to me as a simple analogy:

Would you rather commute on a 256-lane highway or a 320-lane highway? More commuters per second on the 320, speeds being equal. And being in Houston I know you can relate.
This is true to a point... that 320 lane highway, though faster (at the same memory speed) only can only handle half the traffic (1GB vs 2GB). It depends on the size of things this person is rendering.

The bottom line is we cant give a good answer because the OP hasnt listed the model cards for comparison... we need to know that to help out accurately...only one of the links provided go to a card (560ti).
 
Usually the files are around 5gig's for each band. The sound guy will fill up the 32 and 64gig SDHC cards and give them to me to copy and return. I have no idea if he wants to render the 4 or 5 shows at once usually on one of the cards or if he wants to render each band individually. Ideally she wants to get the office guy to start rendering the contents of each card before he leaves and hopefully have the rendering finished in the morning.

The GeForce cards that fall within the budget.

EVGA 012-P3-1571-KR GeForce GTX 570 HD
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130613

EVGA 012-P3-1571-AR GeForce GTX 570 HD
http://www.newegg.com/Product/Product.aspx?Item=N82E1681413062

EVGA 02G-P3-1568-KR GeForce GTX 560 Ti (Fermi)
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130683

This card bust the budget by $20.00 there is a $20.00 rebate card. Bringing it to $299.00. 320bit with 2gigs of ram.

EVGA 025-P3-1579-AR GeForce GTX 570 (Fermi) HD
http://www.newegg.com/Product/Product.aspx?Item=N82E16814130687

I really have no idea how much time would be gained or lost rendering the video in a comparision between the vairables. Amount of memory, bus speed and Open GL support.

I talked her into upping the budget to get 32gigs of ram versus 16gigs of ram figuring that would be her best bang for the buck on this build. But after doing some research found that Priemer uses Open GL and Cuda cores that allow rendering by the GPU.

Thank you for the reply's and any help or advice.
 
Last edited:
Don't get hung up on the bus width. For CUDA rendering (your primary stated task for this card) you'll want a Fermi based card with as many CUDA cores as possible. CUDA rendering is much more dependent on core count and architecture than memory bandwidth in most cases. The 6xx series doesn't perform as well as Fermi in terms of GPU compute and CUDA performance, so stay away from those. The GTX570 is likely your best option within your budget, BUT ...

Double and triple check that the software you intend to use with the card does, in fact, support the card you're looking to buy. Just because a card does CUDA doesn't mean it'll work automatically out of the box. Adobe products are really picky about which cards can use CUDA acceleration out of the box (modifications to force other cards to work are possible, but might be going too far for you). They restrict it to cards that they've internally validated work properly with their software. Other companies do similar things. Going to a Quadro gets you professional-level driver support and compatibility/acceleration in a lot more programs. That sort of certification and support costs more than I think your budget can support, however.
 
The Quadro cards are out of reach for this build. Another thing I have stumbled across is that pci express 3.0 cards have advantages using Ivy Bridge processors.

Reading the Adobe approved cards and the 570 is on the list, but not the 660. The more I read the bigger the can of worms gets. Any thoughts on how PCI 3.0x16 would help with the 3770k Ivy Bridge processor and a motherboard that is PCI 3.0x16?

Thank you for the help guys. Off to go do more research. A googlin I will go.

Thanks again !!!
 
I would imagine none at all to negligible at best. The 570 doesnt saturate PCIe 2.0 16x bandwidth so a bigger pipe (PCIe 3.0 16x) wont help.
 
Back