Timeframe for Kepler?

[SamuelAB]

I'm really interested iray support for Kepler based graphic cards. I was wondering if there is any kind of timeline for this to happen? (2012-2013).

I'm sure many of us are anxious to upgrade to new top of the line system (Macbook pros, Lenovo W330, Dells Precisions, new GT cards...), but do not want to lose the iray feature in the process.

[steve]

I'm really interested iray support for Kepler based graphic cards.

We are currently focused on making this happen as soon as possible. It appears to be a bigger step than anticipated. Going to Kepler is a major shift in GPU architecture, so iray (and other GPU algorithms) need to be adjusted to take full benefit of this. At the same time, a new CUDA version and compiler, plus a new driver generation, will be incorporated, having quite some impact on the overall performance.

We are expecting to get iray running smoothly on Kepler later this year.

Best, Steve.

[SunDevilDave]

We are about to refresh our farm. The original quote had 7 Kepler K10s but I am not excited about getting 1st Gen Kepler cards that we can't use for an indefinite amount of time AND that are hamstrung in terms of double precision. I am going to advise that we go with Fermi based cards instead. It kills me that we won't have that added performance of Kepler, but we need to render.

We specifically went with a render nodes that will be able to upgrade to Keplers...the IBM people told us that iDataplex is going to be forward compatable. BladeCenter expansion GPUs don't have any Kepler options presently.

I go back and forth on this one. Our farm probably won't be in place until August-September timeframe. I need to decide soon though to get it through purchasing.

Edit: We are going with Tesla K10 as of today...if someone has a legit objection please speak up!

[SamuelAB]

Yep, I'm hoping to buy a Lenovo W530 laptop for school. As a student, this is a huge investment and I want to make sure I can actually use the darn thing if I blow 2300$ on it.

I'll most likely wait till next year.. Thanks for the reply!

[rawalanche]

What is the estimated performance increase of GTX680 compared to it's predecessor (GTX580)?

[SunDevilDave]

What is the estimated performance increase of GTX680 compared to it's predecessor (GTX580)?

Hard to say. Its not going to cut render times to a third...but as they learn to write to the new hardware...who knows. I would hope for an initial 30%-50% performance increase.

[SamuelAB]

What is the estimated performance increase of GTX680 compared to it's predecessor (GTX580)?

Have a look at the specs: 680 vs 580.

With 1536 CUDA cores vs 512, you're bound to see a difference, even if they end up being less powerful. Don't forget the 690 will have 3072 cores...... It's a promising future.

I also saw the Intel Phi come up, that look interesting. I'm hoping it gets some application support for rendering at some point if the price is under 1k$.

[rawalanche]

So far from the tests i have seen 680 turned out to be slightly slower than GTX 580 (In GPU rendering tests). Although it could change with iRay being optimized for Kepler architecture and new CUDA version, ARC (Mental Images) have a long history of failure, so I am not being very optimistic here. Performance of MentalRay is same for years (if not worse), performance of new iRay 2.0 doesn't seem to be any different at all, so i really do not believe there is anyone competent enough to pull of some miracle. My best bet is maybe 5-10% increase compared to GTX 580 And that is way below paying off.

[SamuelAB]

Kepler does not work at all with iray at the moment as far as I know. As well, I'm not sure where you are getting your results, but I'm guessing it's from video games, which won't take much advantage of the parallel processing. Would you mind posting your sources?

I'll bet you 5$ we'll see at least 50% performance improvement within a year of Kepler support on the 680 vs the 580.

[rawalanche]

Kepler does not work at all with iray at the moment as far as I know. As well, I'm not sure where you are getting your results, but I'm guessing it's from video games, which won't take much advantage of the parallel processing. Would you mind posting your sources?

I'll bet you 5$ we'll see at least 50% performance improvement within a year of Kepler support on the 680 vs the 580.

http://www.anandtech.com/show/5699/n...-680-review/17