Kepler test build 2.58e Standalone (for GTX 6xx cards)

[MattRM]

Hi Roeland,

So oki, with alpha channel on, the bug on deep channel go away, but why need alpha channel on an image that not have alpha channel (full geometry on screen, for me now it's a workaround, not a fix, still have something strange. I know, before I fix a bug in a shader as say the same thing to artist ).

For the benchmark scene have used, to have 8192 samples on my previous card (GTX 480 1.5G), it take 32min, here 41 min . And on the GTX 480, there was, if I remember well, 450 CUDA unit@1400Mhz. I know that NVIDIA choose to down the GPU compute speed to avoid great heat on gpu (and that's right no high temperature and a real silence card), but 1344 cuda vs 450... speed should be a few more than GTX 480 (more cuda but less MHz, seem to be more fast as well or quite the same). I know, it's a concurant, but in Indigo, in hybrid mode, I get more sample than previous without have a real slow-down of the desktop (a real real slow-down, very hard to click on something to stop rendering or move the mouse). For the speed, in idle, normally it's the boost speed that run, and normally depending of the temperature of the core, run beetween 928Mhz and 1006.

Roeland, if you can say something also about my last post about idea and co, and tell us where you go with Octane, it'll be interesting. For now, I need to use Cycle vs Octane to had some extra passes for compositing, but cycle is full integrated inside blender (Octane's plugins for blender ? ), for now less powerfull, but more user friendly (timeline, instances and co). And I know that Brecht was one of yours devellopers. As I said, your product is really good, but the pipeline is quite hard and not really powerfull for production at this time. So could you say me, that is the future of Octane ( I saw you remove from features pages the support of rib, that's the first thing that decide me to buy Octane, cause I know well renderman and have lot of idea for production pipeline ).

Best Regards,

Matt

PS : I know I'm a quite hard sometime, but I do the same with other beta test engine (Guerilla render that's our internal engine, Atomkraft and co but without buying the engine, free access). It's a way to push the product up and I want Octane goes up and up

[marcio_max]

My GTX 670 4GB shows the same thing in Octane. Why is it? 56 Cores?

The number Octane is getting from the CUDA runtime is probably wrong. If Octane starts a render kernel, it always runs on all CUDA cores. The reported speed may be the speed when the GPU is idle. How fast does Octane render the benchmark scene?

For the deep_channel_kernel: switch on the alpha channel.

--
Roeland

Benchmark scene:
Directlightning = 8,53
Pathtracing = 2,45
PMC = 1,74

[MattRM]

Oki, I found the official benchmark scene from Francesco Legrenzi.

So here the result for the GTX 670 4Gb vram :

Direct Lighting to match 8192 sp : 9min32, 8.59 Ms/sec
Path Tracing to match 8192 sp (max depth 16) : 34min20, 2.39 Ms/sec
PMC to match 8192 sp (max depth 16 ) : 49min15, 1.66 Ms/sec

Matt

[marcio_max]

Hi Roeland! Octane Render is creating too much noise in the scene.
It is very time consuming refine them. Is there any update to improve this problem?
Thanx

[MattRM]

Hi all,

I found a free software that read really the data from Zotac GTX 670 4 Giga, this the only one that return real informations.
Here the card performance during benchmark computation. That's right Octane really use 100% of gpu and so enable the gpu in boost mode. The only thing, with mono CG, all the desktop and Octane interactive mode are really slow down like "evening slide". The mouse move with jumping from a point to an opposite point on screen.

Hope can help.

Matt

[MattRM]

@Roeland :

I saw a post for grass instancing in 3dsmax, Is it possible to do this right now with Octane 2.58e ? I'm not afraid to develop a script in python to write out the scene file description, but I need to know the scene file description (for Blender, for me it's not a problem if the script as no interface right now, i'll build it later). I already made python script to build 4500 nuke scenes (with automatic input output directory creation and co), automatic export / convert maya scene to our internal engine on renderfarm... so not really afraid to output a text file scene description (python support multithreading so...). But I not know C or C++, so not integrated plugins.

Of course will be better to have a blender script like 3dsmax or maya. I think will be a good idea to have it at top list in the plugins list dev. Blender grow up and grow up (including in studio, that's why at start to learn Blender).

Best regards,

Matt

ps : of course if it's possible to connect already existing macro material in file description or other I take it too

[roeland]

No, 2.58e doesn't support instancing. When the instancing is done, the next thing is implementing a way to export instances to Octane.

--
Roeland

[MattRM]

Oki, so the plugins support instances, but not the standalone ? or I missed something ? Cause in this way, plugins is not the same engine like the standalone (our current render engine in the studio are little differences than renderfarm, cause interface is include in the engine, relighting system too). Do you planed to modernise the scene description for the standalone, like change data to not use at all obj format and enabled instances and maybe support other features like I wrote few days ago ?

Best regards,

Matt

[vladislav]

buy now 680 4 GB. tested ....... I'm shocked it's slower 470.
I'm waiting for you to optimize for Kepler.

[marcio_max]

Hi Roeland! Octane Render is creating too much noise in the scene.
It is very time consuming refine them. Is there any update to improve this problem?
Thanx

I found the solution: PMC! Is the best way!