Point instancing running out of memory

[andrewprousalis]

Running into a problem (bug?) that has cropped up sometime between 2023.1.0.2 and 2023.1.3. I used to be able to render 20 million point instances with relative ease, and now the system runs out of system memory and fails. I have built an example scene for those that are interested.

Error on the console after several minutes of loading onto the gpu:
*** OCTANE API ERROR: CUDA error (1) on device 1: invalid argument
*** OCTANE API ERROR: -> failed to bind texture to device memory(0:__data4__)
*** OCTANE API ERROR: device 1: failed to bind data texture '0:__data4__'
*** OCTANE API ERROR: CUDA error (1) on device 0: invalid argument
*** OCTANE API ERROR: -> failed to bind texture to device memory(0:__data4__)
*** OCTANE API ERROR: device 0: failed to bind data texture '0:__data4__'

I can render 20M points rather quickly, its just when they are instances it fails.

IPR mem usage readout

houdini_zxGDL2jJWP.png (17.31 KiB) Viewed 1644 times

[ricky_otoy]

Running into a problem (bug?) that has cropped up sometime between 2023.1.0.2 and 2023.1.3. I used to be able to render 20 million point instances with relative ease, and now the system runs out of system memory and fails. I have built an example scene for those that are interested.

Error on the console after several minutes of loading onto the gpu:
*** OCTANE API ERROR: CUDA error (1) on device 1: invalid argument
*** OCTANE API ERROR: -> failed to bind texture to device memory(0:__data4__)
*** OCTANE API ERROR: device 1: failed to bind data texture '0:__data4__'
*** OCTANE API ERROR: CUDA error (1) on device 0: invalid argument
*** OCTANE API ERROR: -> failed to bind texture to device memory(0:__data4__)
*** OCTANE API ERROR: device 0: failed to bind data texture '0:__data4__'

I can render 20M points rather quickly, its just when they are instances it fails.

houdini_zxGDL2jJWP.png

Hi there, yes please could you give us a scene file.

What GPU are you using, is it in win10 or 11?

There is an Nvidia driver issue at the moment, please try: https://www.nvidia.com/download/driverR ... 156/en-us/

Thanks!

[andrewprousalis]

This workstation has dual 3090s on windows 10, but behavior is consistent across multiple machines, including some single 3090 and dual 4090 machines. That driver (552.22 Studio) is the one I'm currently on. Here's the scene file, thanks for the help!

[ricky_otoy]

I am on win11 and have tested out on a few builds (including 2024) and can't reproduce this issue.

This will be a core (non-plugin) issue, but I still think it is most likely driver related. I am personally on 546.17 (non-studio)
** Please check the scene ORBX exported to Standalone and see if the failure occurs there too.

Things to try:

• On a dual card setup, change which card does the imager/tonemapping.
  Disable any denoise
  See if HAGS is enabled on your win10 machine.

HAGS

• 1) Open Start Menu and click on Settings.
  2) In Settings, go to ‘System’ and open the ‘Display’ tab. Under the “Multiple Displays” section, select “Graphics settings”.
  3) Turn on or off the “Hardware-accelerated GPU scheduling” option.
  4) Restart the system