[FEATURE] Support rendering on GPU

[rabits]

Is your feature request related to a problem? Please describe.

Since now local provider is supported - we can work on GPU support. Video cards (like RTX) could be really fast in e-cycles for example.

Describe the solution you'd like

Clouds are supporting GPU, local support GPU and Manager should have configs to choose the rendering device.

[rabits]

I think current workaround will be to not specify CPU in https://github.com/state-of-the-art/BlendNet/blob/master/BlendNet/script-render.py#L50 . GPU usage also requires additional configuration in the system Blender settings which is not supported by the cloud providers (GCP/AWS), but for Local provider you can setup the environment as you want to use your own blender settings.

[slatunje]

My main focus right now is to render via GPU in GCP asap, ideally using blendnet.

So I took it upon myself to spike simple code changes to get gpu implementation working within blendnet.

Issues to look at and explore:

• allow addon ui to configure linux distro
  • project = 'ubuntu-os-cloud'
  • family = 'ubuntu-2004-lts'
• allow addon ui to configure gpu details
  • accelerator_type = 'nvidia-tesla-p4'
  • accelerator_count = 1
• allow addon ui to load custom startup script
  • startup_script = /path/to/custom_script
• allow addon ui to display list of available accelerator types based on selected pre-configure zones
  • user will add list of preferred zones, gpu count
  • addons uses zones and gpu count to return list of accelerators
  • user selects the accelerators
    • addons validates GPU => vCPU => Zones valid for P4 only available in specific zones, with only 1-24 vCPUs
• allow scene to select GPU if configuration is valid
• cuda drivers take so long to install, so next time I will explore the possibility of using custom ubuntu image and/or machine type.

These are just things that came to mind whilst spiking.

[slatunje]

@rabits #118

Also tested with custom image and it works very well.

Do bear in mind that https://github.com/state-of-the-art/BlendNet/blob/master/BlendNet/script-render.py#L50 is hard coded to GPU

The issue to resolve right now is to decide:

• what values will be read from config file
• what default and custom start scripts read from file paths
• some values needs to be read from ui e.g GPU/CPU