OctaneRender™ Standalone v1.0 beta3.02x [experimental]

A forum where development builds are posted for testing by the community.
Forum rules
NOTE: The software in this forum is not %100 reliable, they are development builds and are meant for testing by experienced octane users. If you are a new octane user, we recommend to use the current stable release from the 'Commercial Product News & Releases' forum.
Post Reply
User avatar
abstrax
OctaneRender Team
Posts: 5509
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

Hi all,

We received some reports about bad performance of the beta 3.02 releases. Beta 3.02 includes many changes, with a new BRDF and a rewritten CUDA framework being the main ones.

The new CUDA framework gives us more flexibility when we build kernels. To find out how different build options work, we created two versions using the same code, but different build setups. To make comparisons to 3.01 and previous versions easier, these two versions use the old BRDF model, i.e. are not meant for production use (the new BRDF model is much much nicer :)). These versions contain all fixes of 3.02a plus some of 3.03.

When you download the package below and unzip it, it will unpack a folder "OctaneRender_TEST_beta303_<os>" which will contain to sub-folders "x1" and "x2". These contain the two different builds.

We didn't receive any test scenes, which is why I attached 5 of the demo scenes - slightly updated. They all contain now 3 render target nodes, each using a different render kernel. The maximum samples is always set to 1000 and the path tracing and PMC kernels have always a maximum path depth of 8.

The scenes contained in the archive are:
- ChessSet with image textures, for a simple low-poly studio scene
- ChessSet with procedural textures, to test heavy procedural textures
- Hallway, as a very simple indoor scene
- OctaneBenchmark, the (in)famous exteriour trench with only one material (please use octane_benchmark.ocs)
- Screws, a fairly heavy studio scene with some procedurals and heaps of glossy materials

If you want to run the performance tests and report results, we would need the GPU, operating system, driver version and the speeds of one ore more of the test scenes run in beta 3.01, 3.02x1 and 3.02x2. If you are keen you can also render them in older versions and check what performance you get there, but we are mainly interested in the differences between 3.01, 3.02x1 and 3.02x2.

EDIT: If you measure PMC speeds, please wait until at least 100 samples/pixel to have the sampling rate settled.

We would be happy if we would get some test results, especially from those who have performance problems with beta 3.02 or beta 3.03.

Here are the downloads:

Windows 64bit:
http://render.otoy.com/rcdownloads/Octa ... _win64.zip

Mac OS 64bit (updated on 15/10/2012):
http://render.otoy.com/rcdownloads/Octa ... 64_new.zip

Linux 64bit (updated on 15/10/2012):
http://render.otoy.com/rcdownloads/Octa ... new.tar.gz

Test scenes:
http://render.otoy.com/rcdownloads/beta ... scenes.zip

Many thanks in advance and happy testing,
Marcus
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
User avatar
matej
Licensed Customer
Posts: 2083
Joined: Fri Jun 25, 2010 7:54 pm
Location: Slovenia

My results (.txt attached)
bench.png
* 259 is the fastest build
* 301 is the slowest build
* 302 x2 is the fastest among 3xx builds
* From 301 to 302 PMC gains more speedup than other modes
* Screws scene is the only scene where 3xx builds are all faster than 259 (considering margin of error there in PT)
Attachments
bench.txt.zip
(465 Bytes) Downloaded 175 times
SW: Octane 3.05 | Linux Mint 18.1 64bit | Blender 2.78 HW: EVGA GTX 1070 | i5 2500K | 16GB RAM Drivers: 375.26
cgmo.net
User avatar
gabrielefx
Licensed Customer
Posts: 1701
Joined: Wed Sep 28, 2011 2:00 pm

nice...

is it based on Cuda 5.0 toolkit?

I would like to understand now how Octane performs with a GTX 680 4GB...

Abstrax:

you enabled more than 64 textures?
quad Titan Kepler 6GB + quad Titan X Pascal 12GB + quad GTX1080 8GB + dual GTX1080Ti 11GB
User avatar
pixelrush
Licensed Customer
Posts: 1618
Joined: Mon Jan 11, 2010 7:11 pm
Location: Nelson, New Zealand

here you go...one apparent anomaly in 3.03 PMC screws -double checked it though.

my processor is running at 3.16 ghz atm and the 460 is o/c 12.5% core and 6.25% mem.
HTH

edit: updated to include 2.59 - a bit of performance loss since then at least using the same cuda5 driver (although not as much as I was expecting) but the screws scene is better now :) except for PMC :cry:
Attachments
results.JPG
i7-3820 @4.3Ghz | 24gb | Win7pro-64
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55
User avatar
bepeg4d
Octane Guru
Posts: 10356
Joined: Wed Jun 02, 2010 6:02 am
Location: Italy
Contact:

here is my table, unfortunately the mac versions are always slow with some little difference.
the screws scene is crazy in every system :D
i hope that you could find some useful information in it :roll:
light blue for the best score
dark red for strange result
pdf and xls version in the zip file (is made with ipad so i don't know if the xls version is usable)
ciao beppe
Attachments
Or Speed comparison-2.zip
(63.5 KiB) Downloaded 173 times
Or Speed comparison-win.jpg
Or Speed comparison-mac.jpg
User avatar
abstrax
OctaneRender Team
Posts: 5509
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

gabrielefx wrote:nice...

is it based on Cuda 5.0 toolkit?

I would like to understand now how Octane performs with a GTX 680 4GB...

Abstrax:

you enabled more than 64 textures?
No, both test builds are using CUDA 4.2, but you can use them also with Kepler GPUs. I wouldn't expect any differences between 3.02x1 and 3.02x2 though.

The texture limits on Kepler are still the same as on Fermi in Octane. It will be changed soon, but it will be similar to the maximum VRAM you can use on multiple GPUs: The lowest limit of all active GPUs will be the limit.

Cheers,
Marcus
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
User avatar
abstrax
OctaneRender Team
Posts: 5509
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

The reason why the screws scene is so different on 3.02/3.03, is because the glossy materials behave very different in 3.02/3.03.
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
User avatar
pixelrush
Licensed Customer
Posts: 1618
Joined: Mon Jan 11, 2010 7:11 pm
Location: Nelson, New Zealand

What happened to PMC between 3.02 and 3.03? It went from 1.8ish to 1.35 for me - 25%...
i7-3820 @4.3Ghz | 24gb | Win7pro-64
GTS 250 display + 2 x GTX 780 cuda| driver 331.65
Octane v1.55
User avatar
abstrax
OctaneRender Team
Posts: 5509
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

pixelrush wrote:What happened to PMC between 3.02 and 3.03? It went from 1.8ish to 1.35 for me - 25%...
Are you comparing 3.02x (using the old BRDF) with 3.03 (using the new BRDF)? Nothing changed in PMC, but the BRDF changed.

Cheers,
Marcus
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
User avatar
abstrax
OctaneRender Team
Posts: 5509
Joined: Tue May 18, 2010 11:01 am
Location: Auckland, New Zealand

bepeg4d wrote:here is my table, unfortunately the mac versions are always slow with some little difference.
the screws scene is crazy in every system :D
i hope that you could find some useful information in it :roll:
light blue for the best score
dark red for strange result
pdf and xls version in the zip file (is made with ipad so i don't know if the xls version is usable)
ciao beppe
Thanks for the feedback. I think there is something fishy going on on Mac OS. Is the 560Ti officially supported on Mac OS? Which CUDA driver does support it at all? I think the CUDA driver is the problem here, but that's just a wild guess.

Cheers,
Marcus
In theory there is no difference between theory and practice. In practice there is. - Yogi Berra
Post Reply

Return to “Development Build Releases”