OctaneRender® pre-Beta 2.4 (win x64) [OBSOLETE]

[OctaneFX]

Thanks for this awaited release !
I have made my own test of rendering speed with identical scene file :
1min40'' with 500 passes with 2.3v5 = 100 secondes
1min20'' with 500 passes with 2.4 = 80 secondes

Gain = 20% in rendering speed on GTX460 !

[Jaberwocky]

Ok I have the Answer.

For GTX460 owners.

I have tested both the Previous version and this current version using the latest Version of GPU-Z 0.5.

The answer it that both versions used all the cores.The Previous version may have only reported seeing 226 of the 338 cores.However it was using all the cores to render up as both the last version and this version registered around 95-98% usage of the GPU on rendering an identical scene.

Hence why we are only seeing a 10-20% increase in preformance on this version.It's probably down to the optimisation of the code as promised by Radiance.

I think therefore that the question for Radiance and all at Octane is in light of this test if they really want to go through recoding the prgram for Cuda 3.2 or not.

Certainly there seems to be no benefit in spending 6 weeks recoding the open GL code into Cuda 3.2 for little performance Benefit.

The Octane Team certainly do not need to think about issuing multicard and single card versions of the program.They may as well issue the current code still using Cuda 3.1 for everyones benefit.

[ROUBAL]

@Jaberwocky : I think that it is very important to follow Nvidia evolution as close as possible, and code for most recent Cuda versions, because most softwares require to install the last driver, and it is the first thing that asks a software support to do in case of display trouble.

So, not updating to most recent driver for Octane would mean that some other softwares could not work correctly in the future if users have to work with an outdated driver, and for what I know, you can't have two different GPU drivers installed at the same time.

Just my two cents...

[Jaberwocky]

Roubal:

Your probably right in the long term.I was thinking more about the short term and the workload that Radiance is under.Simplifying things for maybe for this release and possibly the next release by just keep using 3.1 and let Radiance sort out some of the other issues then slowly migrate to 3.2 and beyond once the full release is out.Certainly now that we know that using 3.2 will not bring any further perfomance benefits.

Any Comments Radiance ?

[abstrax]

Regarding the funky alpha channel behaviour:
- For compute 2.1 cards (like GTX 450/460/465), please make sure that image width is a multiple of 24.
- For all other cards, please make sure that the image width is a multiple of 8.

We will release an update that takes care of that internally and fixes the problem. (Sorry for the inconvenience)

Cheers,
Marcus

[abstrax]

Ok guys. Thanks for helping us to debug beta 2.4.. I reckon, that's the reason why we call it a beta...

Anyway, now the alpha shadow option works for emitters, too:

Old:

New:

I really thought, I had tested alpha shadows with mesh emitters, but obviously I haven't...

For now we have fixed:
- the too low precision in the roughness fields
- broken alpha channel output at certain resolutions
- broken alpha shadows for mesh emitters

We are currently looking at the cyan emitters. Any other show-stoppers we should fix, before we build for the other platforms?

Again, thanks for your help. Cheers,
Marcus

[joseph]

I hope the 2.4 has the saving bug fixed.......viewtopic.php?f=33&t=4656
also importing mesh (the last button in node inspector) does not keep material settings...hope this will be fixed too....

[abstrax]

I hope the 2.4 has the saving bug fixed.......viewtopic.php?f=33&t=4656
also importing mesh (the last button in node inspector) does not keep material settings...hope this will be fixed too....

Yep, that's fixed in beta 2.4.

Cheers,
Marcus

[GeoPappas]

Jaberwocky:

Which NVIDIA driver versions did you use for your tests? In other words, what driver version did you use for 2.3v5 and which driver version did you use for 2.4?

A lot of people have tested 2.3v5 using the older driver versions and reported that not all cores were being used.

[Jaberwocky]

Abstrax

will you be re re releasing this windows x64 build again along with the other builds once the current list of bugs are sorted ? and if so how will we know ?