New Build GTX 580 3G or GTX680 4GB using Octane render

[Rendez]

I am about to make a new build and want to buy and use Octane Render. I've looked on the forum board and noticed that the GTX 6 series cards use Kepler as compared to the Fermi versions of the GTX 5 series cards which Octane Render utilizes.

Should I buy the GTX 580 3G card (which are getting progressively harder to buy in Australia) or buy the GTX680 4GB card? What are the performance differences with Octane Render for the comparative cards. Also are the GTX 680 cards going to be fully supported and optimized for Octane render soon?

Immediate concern is that I won't be able to source a GTX 580 3G card in the near future as they seem to be getting discontinued.

[FooZe]

Hi Rendez,

With the current Kepler test build the 600 series cards run at about 60-70% the speed of their 500 series equivalents.
I don't want to tread on the dev's toes but here is what i understand the plan and progress to be currently:
1) Full support for both fermi and kepler is planed to be in the coming releases (I believe weeks, not months is the plan here).
2) Optimization has had some progress that brings the kepler a little closer to the fermi performance, so watch out for this in coming releases too.

We can't make any guarantees that the 600 series will out perform the 500 series in octane render so bare that in mind.
Personally I like the 600 series for the lower power usage (less troublesome to stack a bunch in your machine) and the additional VRAM.
Also if you ever want to play games the 680 will out perform a 580.

Cheers
Chris.

[Design-Co]

Hi FooZe:

Your post is from 15th may: There is a new development to those new Keppler Cards?

Thanks and best regards.


DSalasC

[FooZe]

Hi Design-Co,

I think you may have the dates mixed up .
I joined on the 15th may. This post is actually only a week old - 29th august.

Thanks
Chris.

[tglaja]

Hi Rendez,

With the current Kepler test build the 600 series cards run at about 60-70% the speed of their 500 series equivalents.
I don't want to tread on the dev's toes but here is what i understand the plan and progress to be currently:
1) Full support for both fermi and kepler is planed to be in the coming releases (I believe weeks, not months is the plan here).
2) Optimization has had some progress that brings the kepler a little closer to the fermi performance, so watch out for this in coming releases too.

We can't make any guarantees that the 600 series will out perform the 500 series in octane render so bare that in mind.
Personally I like the 600 series for the lower power usage (less troublesome to stack a bunch in your machine) and the additional VRAM.
Also if you ever want to play games the 680 will out perform a 580.

Cheers
Chris.

Are you saying that even after Octane optimization for Kepler, it will still hard to outperform Fermi?
How so?
Isn't Kepler out number the CUDA cores compared to Fermi?

[FooZe]

Hi tglaja,

I'm saying there are no guarantees that kepler will out perform fermi for octane render.
1 fermi cuda core is not directly comparable to 1 kepler cuda core.
They have changed the arrangement of cores and the clock speed they run at...

There are plenty of reviews and articles out there if you want more info:
A quick google will turn up articles like this one, with some info: http://www.anandtech.com/show/5699/nvid ... 0-review/2

Cheers
Chris.

[tglaja]

Hi Chris,

I don't want to sound ignorance or whatever because I don't know anything about the technicality behind it.
I just thought the more the better but it's not as simple as that apparently.

Thanks for the info.

[FooZe]

Hi tglaja,

That's fine
No, unfortunately it's not that simple...
If it were the dev's would have a lot less work to do and straight out of the box the keplers would run faster than fermi but unfortunately this is not the case due to the re-design of the cores.

Thanks
Chris.

[MOSFET]

Nvidia's next release 'Big Kepler' GK-110 (As opposed to the GK-104 GTX 6XX 'Little Kepler' parts) will address CUDA performance. The presentation they gave at Siggraph 2012 stated that comparing Fermi to 'Little Kepler' was an Apples vs. Oranges situation as 'Little Kepler' is a 'much smaller chip' as they put it.

The GTX 780 based on 'Big Kepler' is rumored for release between this December and March 2013 depending on what AMD has to offer. The difference in compute performance as compared to the Tesla K20 will remain to be seen.

[alessandro boncio]

Hi i'm new here and also new with Octane, i have to buy a PC, can you tell me if i can buy a gxt 680 or may is better 2x gtx 580? Thanx in advance,
Alessandro