Kepler test build 2.58e Standalone (for GTX 6xx cards)

[roeland]

Hi all,

We have made an update to our Kepler test build.

This build is made specifically for people who already have a Kepler GPU. If you don't have a Kepler card you should stick to the current commercial version or release candidate.

The performance for now:

• On Kepler cards the performance is better than the first Kepler build, but not yet on par with the Fermi cards. We are expecting we can further improve this performance later.
• On Fermi cards the performance is still below the performance of our normal CUDA 4.0 build.

Also, due to modifications in the material shading code some materials may not render correctly in this build. You can report any errors here.

This version is built for Fermi and Kepler architectures. This means only the Geforce 4xx, 5xx and 6xx cards (and the Fermi based Quadro and Tesla cards) are supported. It uses CUDA 4.2, which is supported by any driver that supports the Kepler cards.

Download: Windows 32-bit — Windows 64-bit

Edit: updated the build to the 2.58e release. The performance is roughly the same as the 2.58c release. Original 2.58c Windows 64-bit build here.

--
Roeland

[PeterCGS]

Does anyone have time to benchmark this? Im looking for new cards with more memory, because Im hitting the roof in every project now. And I rather take mediocre performance now with less powerdraw, than go for the 580 cards... But if its still at par with the 470s, then...

/Regards Peter K

[roeland]

For most scenes the GTX680 with this build seems to render at around 60 to 70% of the speed of the GTX580 with the normal CUDA 4.0 builds.

--
Roeland

[PeterCGS]

Thank you very much for the info Roeland! Do you expect to be able to release improved version of the "Kepler" version during the spring/summer, or will the big bump will be with Cuda5 release in the autumn? Im 50/50 torn between going with a bunch of 690 cards now, but if you think could squeeze better performance in a relative short timeframe (1 month from now), I will jump on it right away for a project Im doing and order a couple of badboys

/Regards Peter K

[ChrisVis]

Hey,
I also want to know, cause I am in the same situation with a job to be done in the next 2 month and the decision to buy GTX680s to be more future proof and see hopefully big performance improvements soon or to buy GTX580s 3GB to get the best performance right now, but be outdated and outperformed with this setup in a couple of month.

Roaland, please give us some insight in the Cuda 4.2 design and the possibilities to get more performance out of it for a Keppler Chip, which design may just be fully accessed by Cuda 5?
It would be fine to see the GTX680 head by head with the GTX580 for now or the next few weeks (not months!), so the investment decision is clear to go for the GTX680 now.

Hope you understand your customers, who need to have work to be done. (and therefore need to make well made investement decisions to have the best performance affordable).
Thanx.

[bb3d]

Nobody can foresee how well Octane will perform on Kepler cards in the future, because CUDA 5 isn't out yet.
As a customer I would buy GTX 5x0 now, because they give you the best bang for the buck and are reliable. Around end of this year the situation will be much clearer and the prices for GTX 6x0 cards will go down a bit.

[tehfailsafe]

Also, is it correct that with kepler number of textures is no longer limited in octane as well?

[roeland]

No, it has been increased, but it is still limited.

--
Roeland

[Zay]

No, it has been increased, but it is still limited.

--
Roeland

Increased to what? Some sites say 96 while others say 128 . Anyone knows what's true?

[roeland]

It has been doubled so if we finish kepler card support we can allow double the amount of textures on those cards.

--
Roeland