Kepler test build [Obsolete]

[roeland]

Hi all,

We have made a build with support for Kepler cards, to find out if and how OctaneRender performs on the new GeForce GTX 680. At the moment we don't have one ourselves, so we can't fine tune the performance yet.

It seems that this build runs very slowly on the GTX680. You can use this build to test if Octane works with this card or not on your computer, but for conclusions about performance you should wait until we support the Kepler in one of our next releases.

As of now you should stick with the commercial version.

This is a build of beta 2.57 against CUDA 4.2 for Fermi and Kepler cards for 64-bit windows. The older GeForce 2xx cards are not supported in this build.

Note also there is still a performance regression compared to CUDA 4.0.

Download: Windows 64-bit

--
Roeland

[PeterCGS]

Exciting!!!! Now where is t_3 when you need him!?

[pixelrush]

So a slow dp card with a slow cuda
Must be trying to make Quadro and Tesla use obligatory.
You know with a smaller die than Fermi they surely could have squashed on another 2 dp units without much cost.
Even if the dp performance was say 3/4 of the Quadro/Tesla or ATI it wouldn't have been so bad.
Be interested to see how the Octane benchmark is.

[t_3]

here i am... bad news: currently it seems to be slower than a gtx 460
will do some more test in a few hours; roeland, if you want me to test sthg, just pm me...

[elwisoroarke]

I hope this isn't a play by Nvidia to force the highest end cards on the GPU graphics crowd...keep testing t_3 & keep us posted. Thanks roeland for the initial 680 specific build to test. <fingers crossed>

[mbetke]

so i better save my bucks for a TESLA or even buy another GPU renderer and use ATI. too bad. lets see what will happen. i have no really hurry for a upgrade to smthg.

[pixelrush]

Slower than a 460
Well I suppose I ought to be OK with that because I have a 460 but that's terrible for a new card that was supposed to deliver big performance gains, 1536 cores etc .
I was expecting about 2.5-3 x faster than mine, so even making some allowance for poor optimisation its got to be only working at half pace.
What have they done to sp cuda?!
If thats how its going to be for the whole Kepler series and going on to Maxwell that's going to kill off buying them for gpu rendering for the masses. Dumb and greedy move from Nvidia IMO.
I wonder how long until Quadro/Tesla come out....surely they won't perform that badly as well.
It seems to me they wouldn't make a whole new die for those, small production run and all, so it seems likely the basic circuitry is the same. That leads me to think the chip might be laser trimmed to kill off some functionality/performance in Geforce.
Perhaps Refractive had better start thinking about openCL and start talking to ATI.
Or...everyone who uses an Nvidia Geforce for cuda apps should complain loudly to them to not cripple their cards excessively like this. No reason why they can't back track their PR blunder and reissue a 680 SE with less trimming
I mean come on it has to have better performance than the last generation or nobody will buy ever buy a new Nvidia card again!

[Timmaigh]

here i am... bad news: currently it seems to be slower than a gtx 460
will do some more test in a few hours; roeland, if you want me to test sthg, just pm me...

that is mightily dissapointing. We shall see, what the fine-tuning does, but i am not holding my breath.

silver lining, at least i dont have to sell my 590 yet

[matej]

Well, if you're gonna cripple your gaming segment cards to favor the "pro" segment, why not do it in a big style

But, seriously, slower then 460? Obviously calling it a 680 is just a marketing trick.

Octane on OpenCL with AMD cards, yes please.

[mbetke]

If only one out of 5 customers switches from a GTX to quadro they still make a profit and give a shit on the other four customers....its easy like that!
SO complaining will do nothing as usual....