NEW GRAPHICS CARD

[megacal]

Thanks for the info...I'll keep it in mind when putting this together....I may yet build one
if I can't find a decent cheap one with 2 available pci express 16 slots.

I was looking at the Palit, but a review said it's heat sink wasn't nearly good
as the Asus. Amazing that there are so many models! I remember when the
4mb Monster card was cool.

Does this look like a winner?
Doesn't have as much video ram, but may do fine.
And as far as modeling and rendering at the same time, I probably wouldn't anyway.

So maybe I just need one slot(?)

[tehfailsafe]

You'll want a second slot regardless, octane uses 100% of the GPU, meaning even using the menus and stuff IN octane are painful without a display card.

You can go with either 2 render cards like the GTX560 and just set it to use only one (the one not plugged into your monitor) or you can get a super cheapo card like an old 9800GT or something just to run the OS and display output while octane runs on the GTX560.

Here's a quick chart I made for cost to value. I used the CUDA Folding @ Home benchmark for the score, then divided the price by the score to get a cost per point. Then for perspective added the costs for having 2 or 3 of each card and that cost, since you ARE limited to the number of cards you can fit in your box!

Code: Select all

card:         score:      price   cost/score     2x score   2x cost      3x score   3x cost
GTX 580         379.7      $500      1.32         759.4      $1000        1137      $1500
GTX 570         342.1      $350      1.02         648.2      $700         1026      $1050
GTX 560 ti      292.8      $250      0.85         585.6      $500         876       $750
GTX 480         329.1      $399      1.21         658.2      $800         987       $1197
GTX 460         218.7      $169      0.77         437.4      $400         654       $507

Though I'm not 100% sure how well the CUDA benchmark score maps across to ms/sec in octane... I would assume it would be near the same curve?

[megacal]

Thanks for the comparison chart!

I don't understand why the 480 rates higher than the 560 (?)

With a single, modern, GPU you can typically expect to see a 1000%-5000% (10X to 50X) speed increase over a typical un-biased, CPU based renderer.

....Octane home page

Do we know what setup was used for the video tutorials?

What kind of speed increase would you expect from a GTX 560?
Even 10x is phenomenal!

How much better is 2gb video ram, i.e. what difference does it make?

[Jaberwocky]

I suspect it's the old Bottleneck Problem with the cut down warp schedulers that has been mentioned in other threads.Both the 460 that i am using and the 560 are based on that same cut down chip.The 480 along with the 570/580 are based on the full fat chip with the increased warp schedulers.Plus the fact that the old 480 had 480 cores, the 560 only has around 380 cores.

[megacal]

Sounds like the 560 should be called a 380 instead

When picking a card, what are the specs that you look for?

I'd never have known the 560 had less cores and a cut down compared
to a 480...seems like tricky marketing.

[Jaberwocky]

Yes the 560TI uses the same chip as the old GTX460. The GF104 with the remaining 8th SM enabled and becomes the new GF114.

For further details see the link.All will be explained

http://www.anandtech.com/show/4135/nvid ... 250-market

The 570 & 580 are a whole different ball game.

[megacal]

Jaberwocky, thanks for the anatech link...big help.

Do you know how much faster Octane would run on a 570 vs a 560?

More cores vs higher clock speed...worth the extra $100 if the difference
in performance (Octane) is ?x better (?)

[tehfailsafe]

I'm currently debating 3 GTX570 for $1000 vs 2 GTX580 for $1000 or 4 GTX560ti for ... $1000...
Any opinions?

[radiance]

If the GTX560's are 2GB models, definately go for those...
I would choose memory over speed, as all of these cards are pretty fast,
and you definately want to render complex scenes with lots of VRAM.

Radiance

[tehfailsafe]

What about 2 GTX580 3gb then?

Speed is a bit of a concern for me on this specific project, I'll be doing hundred of renders...