OctaneRender™ 4 RC 6

[abstrax]

placement node seems to be broken

Here it works. Could you post a scene where it doesn't work? Thanks.

[abstrax]

The Denoiser crashes on two identical machines (MacPro5,1 - 2x GTX 1080), but only if both GPUs are enabled for rendering or if the image resolution exceeds a certain amount.

I was already in contact with Beppe about this issue, but I have a general question concerning this topic:
Would it be somehow possible to render an image sequence first, and then run the Denoiser sometime afterwards? Like some sort of standalone Denoiser that loads previously rendered image sequences?
In this case we could render the images on both GPUs and run the Denoiser only on one GPU.

You can set up Octane to use only one GPU for denoising (Preferences -> Devices...). If a device is used for denoising is completely independent of which device is being used for rendering.

Would it help in your case to switch to only one denoising device?

Regarding the "standalone" denoiser: In principle it would be possible, but a lot of work and we rather would like to have it working directly in the render pipeline. We are working on on getting the memory issues solved for large resolutions. We know what the problem is, but it requires some mid-sized refactoring of the tone-mapping pipeline and thus won't be finished in the next couple of weeks, but hopefully soon.

[abstrax]

So, just talking about upcoming stuff and later on...
As far as render speed will go, are there any more tricks up your sleeves that are going to make rendering even faster, or is it just now compatibility with newer RTX graphic cards that will be the driving force moving forward?

The answer is basically the same since like forever: We don't know because we don't have a crystal ball We still have got various ideas we would like to try, but it's hard to tell in advance which ideas actually improve things. Also be aware that adding more features usually makes things more complex and with that slower rather than faster. We are constantly trying to bring back performance that got lost due to new features, but we can't make any guarantees that this always be the case. We wish it weren't like that, but it is what it is.

[wallace]

Hi Wallace, maybe this issue is also relative : In Universal Material, when coating IOR is 1 this disable transmission (even when coating is supposed to have no effect).

Thanks for reporting, this will be fixed in next version.

Also mixing diffuse and refraction should only be controlled with 1 attribute : the transmission color, not also the albedo color :
When transmission = 1, we can get a completely opaque material just because the albedo color is white :
It's confusing but also unpractical in production (2 different textures needed in 2 different attributes just to layer diffuse and refractive material ..)

Unfortunately this is the only way to implement universal material, due to inter-lobe interactivity. Similar to diffuse reflection + diffuse transmission, whatever that is not reflected is transmitted. Think about this in terms of layers, if the diffuse reflection layer, above the the transmission layer is reflecting 100% of incoming light, then there won't be any light transmitted into the object.

If you want it to be fully transmissive, then the albedo color should be left as black, this is equivalent to saying there won't be any diffuse reflection contributing to the total reflectance, so if the layer below it is transmitting light into the object, then it's transmitting everything into the object.

If this is too conceptually difficult, then it would be best to use specular material / diffuse material.

Also while some unexpected feature are added ( IOR mapping ), some basic material component are still missing :
- Diffuse reflection roughness is missing ( look roughness effect in the Diffuse material )
- Diffuse transmission is missing (look transmission effect in the Diffuse material ) > we would need a "transluscency" attribute to switch from specular transmission to diffuse transmission.

I am aware of the current model doesn't allow specifying the diffuse material's diffuse model and diffuse transmission, it will be added in future versions.

Please use diffuse material for this as of now.

[oscartung]

placement node seems to be broken

Here it works. Could you post a scene where it doesn't work? Thanks.

I think I know what the problem is.
If I click on the the placement node on the nodegraph editor， the gizmo shows and it works by dragging it.
However if I use the object picker or rightclick to pick the object in the render viewport, the gizmo still shows up but when I drag it it does not move.
Is there a way around this?

[calus]

I understand that, but i'm saying : in your implementation the layers are in the "wrong" order, the diffuse reflection layer should not be above the specular transmission layer but below:
Please check again how Disney has extended its principled BRDF to a BSDF :
https://blog.selfshadow.com/publication ... _notes.pdf

And for example a correct implementation of this in Houdini Mantra:
http://www.sidefx.com/docs/houdini/node ... hader.html

[rodrsamuel]

hi! im having a problem on all octane 4 RC this problem persists on RC 6

i work with C4D Octane on large scenes arch viz projects and every time i enable out of core geo/textures i get cuda error 700 ilegal memory access

i have 7x evga ftw3 1080ti Hybrid with 2x evga 1600w power supply to have more than enough power on a ausus x99 e ws usb 3.1 motherboard.
msi on or off doesn't change anything, Gpu and Vram down clock doesnt do any diference. i have 64gb memory in my pc witch at least 50% is free with stock clock of 2133mhz
already unplugged GPUS one by one to check the hardware and the problem persists.
i dont know what else to try, lower parallel samples doesn't solve .
already used all drivers from 391.35

cheers

Thanks for the detailed report. And sorry for not having a helpful answer (yet), but we haven't seen this particular problem here yet and we don't have a setup that is comparable with yours. -> Some more questions first:

Does it happen with specific scenes or all scenes that use out-of-core memory? Do you get the render failure immediately or only after rendering some time? How much out-of-core memory do you use? Does it also happens if you render on less than 7 GPUs, like only 1 or 2 GPUs?

hi, Abstrax thanks for the fast reply it happens even with 1 Gpu connected to the pc, i only work/render with Live viewer out of core is about 1.8GB i have about 30GB free... the project load into GPU and render with no error but as soon i try to change any material i get :

"CUDA error 700 on device 0: an illegal memory access was encountered
-> failed to download symbol(stats data)
device 0: path tracing kernel failed
denoiserThread0: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered
. Restart required
CUDA error 700 on device 0: an illegal memory access was encountered
-> could not get memory info"

so its impossible to work if i cant change/create materials

[wallace]

I understand that, but i'm saying : in your implementation the layers are in the "wrong" order, the diffuse reflection layer should not be above the specular transmission layer but below:
Please check again how Disney has extended its principled BRDF to a BSDF :
https://blog.selfshadow.com/publication ... _notes.pdf

chrome_2018-10-23_08-39-02.png

And for example a correct implementation of this in Houdini Mantra:
http://www.sidefx.com/docs/houdini/node ... hader.html

Hi,

First of all, disney BSDF model is not a 100% de facto standard.

Secondly, it does not really specify which layer is at the bottom or above like we do. It says exactly on your picture that it's a mix between specular BSDF and diffuse BSDF with a specular transmission parameter interpolating.

This means they are doing an equivalent of mix material between specular BSDF and diffuse BSDF.

On the other hand, here we are explicitly specifying that all transmission layers will be below reflection layers, because they only fall below transmission layers if they are not reflected. There is really no 100% correct way and incorrect way in implementing a union of BSDFs, which is basically what the uber material/universal material is.

This approach is necessary because our transmission will not only specify specular transmission, but also diffuse transmission (this is coming like you have wanted, with a toggle switching between modes), so the equivalent subtraction to get your material transmission makes perfect sense and consistent for both cases. Let's say if you have a rough specular transmission below your diffuse reflection, you should get approximately similar results as you would do with diffuse transmission and diffuse reflection, because the similar amount of radiance gets through into the material in both cases, if they are not reflected outwards in the diffuse layer.

Cheers,
Wallace

[calus]

First of all, disney BSDF model is not a 100% de facto standard.

Secondly, it does not really specify which layer is at the bottom or above like we do. It says exactly on your picture that it's a mix between specular BSDF and diffuse BSDF with a specular transmission parameter interpolating.

This means they are doing an equivalent of mix material between specular BSDF and diffuse BSDF.

On the other hand, here we are explicitly specifying that all transmission layers will be below reflection layers, because they only fall below transmission layers if they are not reflected. There is really no 100% correct way and incorrect way in implementing a union of BSDFs, which is basically what the uber material/universal material is.

This approach is necessary because our transmission will not only specify specular transmission, but also diffuse transmission (this is coming like you have wanted, with a toggle switching between modes), so the equivalent subtraction to get your material transmission makes perfect sense and consistent for both cases. Let's say if you have a rough specular transmission below your diffuse reflection, you should get approximately similar results as you would do with diffuse transmission and diffuse reflection, because the similar amount of radiance gets through into the material in both cases, if they are not reflected outwards in the diffuse layer.

Ok but do you know any other renderer doing this the same way as in the universal material ?
All production renderers have Ubermaterials and to my knowledge none of them have implmented the diffuse Albedo as a a layer above transmission, there must be a reason...

Anyway my main concern is about usability, Ubershaders are often used in production to have one material for one object even if the object have several different sub-material part, so there must be a simple way in the material itself (so only one texture in one parameter) to have a glass part and a red plastic part for example.
So in the universal material, instead of only plugin a mask in transmition I also need to plug in the albedo the inversed mask multiplied with the diffuse color ... far from production friendly ....

If I insist on this point this is because we are next to the official release and I'm afraid after that the universal material implementation will never change for compatibility reason, so this was the time to tell my doubts.

[wallace]

Anyway my main concern is about usability, Ubershaders are often used in production to have one material for one object even if the object have several different sub-material part, so there must be a simple way in the material itself (so only one texture in one parameter) to have a glass part and a red plastic part for example.
So in the universal material, instead of only plugin a mask in transmition I also need to plug in the albedo the inversed mask multiplied with the diffuse color ... far from production friendly ....

If I insist on this point this is because we are next to the official release and I'm afraid after that the universal material implementation will never change for compatibility reason, so this was the time to tell my doubts.

OK, I think you have a point in masking. How about this?

Basically I can extend the universal material later by adding a "Transmissiveness" / "Transmission Map" (or some other name) to it after stable, if it is specified, then it would use the mask value to determine the degree of transmission (the disney way), if not then it would use the current way (the transmitted energy is those that is not reflected off in upper layers).