Smarty wrote:
poky patrol wrote:

On Mon, 31 May 2010 19:27:27 +0000 (UTC), "Smarty"
wrote:

I am using an nVidia GeForce 8800 GT video card purchased about 2
years ago and have several programs which speed up rendering and
filtering by using the CUDA GPU.

I am wondering if a new video card would significantly improve my
CUDA performance? Since many of the rendering and filtering jobs I
do take a long time (several hours per job) on this Intel Extreme
QX9650 quadcore machine, I am trying to find a way to significantly
shorten the renders and filtering.

I have 1 PCIe X16 slot on this Dell XPS420 mainboard, so this may
limit my choices as to which, if any, nVidia card could be used as
an update.

I very much appreciate any suggestions / comments.

Yes, a new top of the line card will give you better performance for
sure but you are looking at spending over $500.00 to get it. If your
mb has SLI I would just put in a second 8800GT instead and that will
give you better performance for less cost. But that depends on if you
have a motherboard with dual PCI-E slots or not and if it supports
Nvidia SLI and not ATI Crossfire.

Thanks for your reply!

My Dell motherboard does not support SLI, and only has a single 16X
PCI-E slot. I guess this means I am left with only an expensive single
card replacement? Is the $400 upgrade the "ONLY" way to get a GPU
improvement for CUDA, and if so, what type of speed improvement am I
likely to see compared to my older 8800GT?

Thanks once again.


I'd want to see some benchmarks first, before spending one thin dime on
new hardware.

Not very many algorithms are "perfect scaling", able to use as many
computing elements as you can give them.

The best place to get this kind of information, is from the company
that sold you the software. And that is one of my pet peeves, that
such information is virtually non-existent. How are end-users supposed
to know how to get the best from their software ? The companies
writing the software, sure aren't helping...

It would be pretty sad, if you bought a GTX480, ran the algorithm,
and it didn't run any faster than it did on your 8800GT.

Paul