OctaneRender® for 3ds max® v2020.1-RC2 - 10.06 TEST

[mbetke]

I have problem with NVLink:

I applied the bridge to my two Titan RTX cards (24GB each). Then I activated SLI. I have Nvidia Studio Drivers 441.61. Win 10 tells me about a total of 90GB GPU memory. This is strange because it should be 48GB.

Well in Octane I put up a scene renders it but get a "CUDA error 2 on device 1: out of memory -> failed to allocate device memory" like without NVLink. The scene was about to render with less vegetation. So the error is about memory.

In the other threads here I read about an extra memory bar which is about to appear when the memory gets shared to the other GPU. But Octane just behaves like without NVLink.
It also doesn't matter if I tick out of core memory.

Whats wrong here?

[samsue]

have a slave for this version?

[paride4331]

have a slave for this version?

Hi samsue,
when the stable version will be released.
Regards
Paride

[paride4331]

I have problem with NVLink:

I applied the bridge to my two Titan RTX cards (24GB each). Then I activated SLI. I have Nvidia Studio Drivers 441.61. Win 10 tells me about a total of 90GB GPU memory. This is strange because it should be 48GB.

Well in Octane I put up a scene renders it but get a "CUDA error 2 on device 1: out of memory -> failed to allocate device memory" like without NVLink. The scene was about to render with less vegetation. So the error is about memory.

In the other threads here I read about an extra memory bar which is about to appear when the memory gets shared to the other GPU. But Octane just behaves like without NVLink.
It also doesn't matter if I tick out of core memory.

Whats wrong here?

Hi mbetke,
could you try installing this Nvidia driver?
https://www.nvidia.com/download/driverR ... 7537/en-us
Set out of core around 64GB and don't forget Nvlink works with textures only.
Regards
Paride

[paride4331]

Started logging on 02.04.20 04:37:13

OctaneRender Enterprise 2020.1 RC2 (8000004)

CUDA error 2 on device 0: out of memory
-> failed to allocate device memory
device 0: failed to allocate post-fx buffer
denoiserThread0 : Not enough free memory to run the denoiser. Required memory 453.47 MB
denoiserThread0: CUDA_ERROR_OUT_OF_MEMORY: out of memory
. Restart required

Hi Elvissuperstar007,
viewtopic.php?f=80&t=74143&p=380170#p380170
Regards
Paride

[samsue]

I have problem with NVLink:

I applied the bridge to my two Titan RTX cards (24GB each). Then I activated SLI. I have Nvidia Studio Drivers 441.61. Win 10 tells me about a total of 90GB GPU memory. This is strange because it should be 48GB.

Well in Octane I put up a scene renders it but get a "CUDA error 2 on device 1: out of memory -> failed to allocate device memory" like without NVLink. The scene was about to render with less vegetation. So the error is about memory.

In the other threads here I read about an extra memory bar which is about to appear when the memory gets shared to the other GPU. But Octane just behaves like without NVLink.
It also doesn't matter if I tick out of core memory.

Whats wrong here?

Hi mbetke,
could you try installing this Nvidia driver?
https://www.nvidia.com/download/driverR ... 7537/en-us
Set out of core around 64GB and don't forget Nvlink works with textures only.
Regards
Paride

for the Titan RTX cards the octane combines the vram? For 2080Ti too?

[mbetke]

Hi mbetke,
could you try installing this Nvidia driver?
https://www.nvidia.com/download/driverR ... 7537/en-us
Set out of core around 64GB and don't forget Nvlink works with textures only.
Regards
Paride

Paride, it just doesn't work in production for me. I updated drivers and set to out of core to 64GB. My scene is nothing special. Around 44Mio tris in whole scene with lots of small geometries objects and textures. Some trees and such stuff. Just arch-viz. No large 150Mio tris mesh or other fancy stuff.
Denoiser also kills the rendering with memory allocation error.
How can I analyze further? This is really annoying to have a top-notch GPU rig here and the announced feature just don't work and I also have no possibility to check what's wrong.

EDIT: Sometimes it stopped on same scene with 0.1GB with denoiser. Without denoiser it works and extends 2.5GB. Really strange that same scene sometimes works and sometimes not. All static. No simulations/animations in there.

Also gets killed when just changing exposure dirung render.

Code: Select all

CUDA error 2 on device 0: out of memory
  -> failed to allocate device memory
denoiserThread0 : Not enough free memory to run the denoiser. Required memory 604.92 MB
CUDA error 2 on device 0: out of memory
  -> failed to allocate device memory
denoiserThread0 : Not enough free memory to run the denoiser. Required memory 604.92 MB
denoiserThread0 : Initialization failed. Restart required
CUDA error 2 on device 1: out of memory
  -> failed to allocate device memory
OptiX error for device 1: [ERROR] "sbt->raygenRecord" is null
OptiX API error for device 1: Invalid value
device 1: path tracing kernel failed
Paride

[Rik]

try changing which card does the denoising?

I had to put 1 card on render duty and another on denoising.

Actually I now have 2 2080ti doing the rendering and 1 old 10606gb doing the denoise. Working great so far...

[mbetke]

I use this configuration. One denoises. The other ones tones. Both render.

[Rik]

Apart from it being slower, does it work if you untick the active box for the card which does the denoise? So 1 card does the rendering, the other does the denoising?