OctaneRender™ Standalone v1.0 beta3.02a [obsolete]

[roeland]

The OctaneRender™ Team is proud to announce the test release of OctaneRender™ beta 3.02a.

This version is compatible with all versions of the NVIDIA drivers that are compatible with CUDA 4.2.
There appear to be performance issues with the most recent drivers.

Please read the release notes of v1.0 beta3.02. There is a file format change, scenes made
in this version of Octane will not be compatible with previous versions of Octane.

Changes since beta 3.02

• Texture mapping works correctly with the 4.2 versions of the NVIDIA CUDA driver.
• Fixed bugs in clay mode rendering causing block-like artifacts.
• IES files can be dropped onto the node inspector.
• Fix loading of embedded IES files in macros.
• When importing OBJ mesh files, Octane creates material pins for material names
  that not present in the OBJ file but not in MTL file. The material pins are also
  created if the MTL file is missing.

Changes since beta 3.01

• The file format has been updated. Older versions of OctaneRender can't load scenes saved in this version.
• The glossy BRDF has been updated. The roughness control has a more linear behavior, and the Fresnel effect
  can be controlled better.
• CUDA Framework update: There is no longer a separate Kepler build, and users don't have to install a CUDA
  toolkit anymore on Mac OS and Linux.
• Color accuracy has been improved compared to previous versions.

Download here:

Windows:
ZIP archives:
- 64-bit — 32-bit

MacOS X:
- 32-bit and 64-bit

Linux:
- 64-bit

Driver Requirements
This build requires an NVIDIA driver supporting CUDA 4.2.

For Mac OS you need a CUDA driver. We recommend using version 4.2.10 (http://www.nvidia.com/object/mac-driver-archive.html).
The display driver should be automatically updated by Mac OS.

On Linux and Windows you only need a recent enough NVIDIA driver. There appear
to be some performance issues with the latest drivers, the recommended driver version
is 301.42 (http://www.nvidia.com/Download/Find.aspx?lang=en-us).


Most recent documentation: 2.58e user manual

Yours,
The OctaneRender™ Team.

[j7th]

Thank you

[abstrax]

Today we found a fix for the texture mapping issue, i.e. you can use this build with CUDA drivers that are older, as long as they are supporting CUDA 4.2. On Windows NVIDIA driver 301.42 works well.

We did some tests to see how different build options impact performance and on the cards tested (GTX 590/690), and we couldn't gain any improvement, i.e. the options used for beta 3.02 worked best and we used the same setup for beta 3.02a.

What made a difference though was switching back to the second last stable NVIDIA driver (301.42), which didn't change much on the GTX 590, but made the GTX 690 performance consistent (i.e. you immediately get the best speed using path tracing/direct lighting).

We still plan to build a version with the old BRDF model to verify that any drop in samples/second is caused by the BRDF changes.

One last thing, some may have already recognized: If you use glossy/specular materials a lot, the noise level after the same amount of samples/pixel is usually quite a bit lower with beta 3.02 than compared to older versions. -> Performance is not only samples/second but actual image quality after the same render time.

Cheers,
Marcus

[grimm]

Yay!! This one worked. Except with the old test ocs files, although the benchmark ocs worked. Must be newer. I did notice that this beta was less responsive than the previous versions on my single GPU system. Happy Jason, now I can have some fun with instancing. If I upgrade CUDA to the 4.2 version will that make any difference?

Thanks,

Jason

[abstrax]

Yay!! This one worked. Except with the old test ocs files, although the benchmark ocs worked. Must be newer.

Which scenes did not work and how did they not work?

I did notice that this beta was less responsive than the previous versions on my single GPU system. Happy Jason, now I can have some fun with instancing.

Hmm, could you please be more specific? What does "less responsive" mean and against which version are you comparing?

If I upgrade CUDA to the 4.2 version will that make any difference?

I guess you mean the CUDA toolkit. No, the CUDA toolkit is not used/required anymore.

Thanks,

Jason

Thanks a lot for your feedback,
Marcus

[Rickky]

any chance to show us a test scene with the index/fresnel effect.
Are you sure it works ?
Ooops, I get it now

[Jaberwocky]

I've run another speed test on this version.

I've taken a furnishing model and added procedural textures and non procedural textures to it.

In each case, using a procedural texture reduces the render speed by as much as 3 Ms/sec.So, using 2 procedural materials on the scene instead of 2 materials using jpg maps, reduces the scene by as much as 6 Ms/Sec.

This is just a word of warning for anybody using procedural textures in a scene,should anybody need to get the the maximum speed out of their cards.

[bepeg4d]

olà.
another speed table comparison between 3.01 and 3.02a.
i have used almost all the demo scenes, the yellow ones are personal test scene (i have asked to my client if i could share but the answer was no)
now in win7 is much better with a boost in almost all the demo scene exept for the Screews that gives strange results. In my scenes, an interior full of IES lights and a complex specular object, the results are a bit discordant but maybe is necessary to rebuild the scene from scratch for a correct comparison
on the mac side always bad results, i have tryed to downgrade the cuda drivers to 4 with different version but nothing, under 10.7.5 only with cuda 5.06 the gtx5xx are recognised.
ciao beppe

[gabrielefx]

good test!

the new Octane version is promising

[abstrax]

olà.
another speed table comparison between 3.01 and 3.02a.
i have used almost all the demo scenes, the yellow ones are personal test scene (i have asked to my client if i could share but the answer was no)
now in win7 is much better with a boost in almost all the demo scene exept for the Screews that gives strange results. In my scenes, an interior full of IES lights and a complex specular object, the results are a bit discordant but maybe is necessary to rebuild the scene from scratch for a correct comparison
on the mac side always bad results, i have tryed to downgrade the cuda drivers to 4 with different version but nothing, under 10.7.5 only with cuda 5.06 the gtx5xx are recognised.
ciao beppe

Wow thanks for the effort and feedback. On the Mac side: Did you use the very latest CUDA driver 5.0.36?

There definitely seem to be some issue on Macs and we will try get it sorted although it might take some time. I hope the planned performance test builds will shed some light on it.

Thanks again for your help,
Marcus