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

[A2Zen]

to compare with VRay when we put an index value > 1

Specular color in OCTANE = Fresnel color in VRay ?

[abstrax]

I edited my previous post with additional info.

Could you set up a similar render target, using PMC instead of PT, then switch to this render target and then back? How does it behave? Does the path tracing run slower, as fast or faster than before?

I hope I did it right: I started with PT, then changed to PMC, then back to PT - in the same (default) render target, and by creating two separate render target nodes. In bot chases PT runs exactly the same after the switch.
(so no speed-up "bug" for me )

Thanks for the feedback.

Regarding instancing/non-instancing: The benchmark scene is obviously a fairly simple scene, i.e. it's a lot more bound to ray tracing performance than your usual scene. Which is why it's actually a bad benchmark scene. Since the slow down happens mainly in the ray tracing part, this scene is more affected than others. Usually the slow down is more around 10%.

We may try get this speed back by building special versions of the kernels with everything regarding instances being removed, but we need to investigate first if it's feasible or not. And even if it is it might take some time since it's very likely a non-trivial change introducing a lot of noise into the code.

Cheers,
Marcus

[matej]

Another benchmark, with a custom project ("product-viz" open scene, no instances, lots of polygons & textures, hdri lighting).

PT

* 259 -> 6.10 Ms/sec
* 302 -> 4.90 Ms/sec (20% slower)

PMC

* 259 -> 2.08 Ms/sec
* 302 -> 2.07 Ms/sec (same speed in this case)

DL

* 259 -> 10.34 Ms/sec
* 302 -> 8.10 Ms/sec (22% slower)

Interestingly PMC doesn't show any slowdown in this particular scene. But in another one (similarly heavy on geometry & textures, a lot of specular material) it showed 12% speed drop. PMC seems to have the smallest percentage of speed drop among all the rendering modes.

Another thing: The loading times of heavy scenes is noticeably longer in 302 than in 259. The process actually freezes (CPU usage goes up, animated cursor icon freezes, the window becomes unresponsive (but the system & other programs work ok)) several times during the "analyzing obj" / "parsing face" / "parsing vertex" phases of loading. It manages to successfully load the scene, but it takes 2x - 3x more time than 259.

I dunno if this is my system or a Linux thing, so if anyone else notices this, please confirm.

[kavorka]

Another benchmark, with a custom project ("product-viz" open scene, no instances, lots of polygons & textures, hdri lighting).

PT

* 259 -> 6.10 Ms/sec
* 302 -> 4.90 Ms/sec (20% slower)

PMC

* 259 -> 2.08 Ms/sec
* 302 -> 2.07 Ms/sec (same speed in this case)

Maybe this is the bug Marcus talked about creeping in? Did you try switching back and forth for this particular case? Maybe it doesnt happen every time since you didnt have it happen previously.

[matej]

I tried it again and I confirm those numbers, PMC speed is the same in 259 and 302 (for this particular scene). Changing back & forth between PMC & PT in 302, doesn't change their render speed.

[marcio_max]

Yeah! Much better!

[pixelrush]

20% is a fair hit. I'm thinking should I change or not? I have to mod the driver as well
Is this performance fixable in 3.03? or is this a consequence of a unified build and including instances?
I don't mind a 10% penalty for the convenience of having everything together but 20% is getting fairly hungry.
Perhaps I should wait to see some Windows numbers first...

[matej]

@marcio_max, you are using a Kepler card, right?

It seems the new build is slower on Fermi, but faster on Kepler, compared to the old ones. Now we would need an speed comparison between say an 580 and 680 (preferably on the same OS).

I bought a 570 just today. I hope it wont turn out to be a mistake...

[kavorka]

@Marcus: Maybe we should get a scene for benchmark that's an interior scene instead of the current benchmark scene?
That way we can all test the same scene and use a scene that is more complex than the current one.

[grimm]

Unfortunately it still doesn't work on my system. I have updated to the latest driver (304.51). Octane loads and displays fine but when I try to render anything (like the spaceships) it spews out several errors:

Code: Select all

[jason@gemong OctaneRender_TEST_beta302_linux]$ ./octane 
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Kernel execution failed (dl)
CUDA device 0: Direct lighting failed
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to unbind texture (data6)
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to copy memory to device.
CUDA device 0: Failed to load data of data texture 5 onto device
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to deallocate device memory
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to unbind texture (data14)
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to copy memory to device.
CUDA device 0: Failed to load data of data texture 13 onto device
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to deallocate device memory
CUDA error 700 on device 0: An exception occurred on the device while executing a kernel. The context cannot be used anymore and must be destroyed. All existing device memory allocations from this context are invalid and must be reconstructed.
    -> Failed to launch kernel (clear)

I tried several kernels and none of them worked. Here is a snap of my preferences, looks like the memory is not being detected correctly for my card.



Thanks,

Jason