ā† Home

šŸ’„ Blender render farm on Google Cloud Platform

! This actually cost more than I would have liked. A few dollars to complete this guide. As a result I donā€™t recommnd following it, but Iā€™ll leave it up in case you want to see what I did. !

I wanted to do some blender renders in the cloud and came across this helpful article on Medium that suggested using Google Cloud Platform and listed some compelling benchmark costs.

Here is a more updated version of that article.

Provisioning

  1. Type quota in the search bar and go to All Quotas
  2. Filter for gpus_all_regions
  3. Request and increase if it is 0
  4. Follow the linked article above to provision a GPU VM

Even with quota raised, you might still have trouble obtaining a GPU instance in your desired region because of high demand.

I got lucky obtaining a V100 GPU in asia-east1.

VM setup

These steps summarise this Google article BUT also make sure Cloud API access scopes is set to Allow full access to all Cloud APIs so that we can upload render output to a GCloud bucket and download our renders.

  1. SSH into the VM and runs these commands to install nvidia drivers:
curl -L https://github.com/GoogleCloudPlatform/compute-gpu-installation/releases/download/cuda-installer-v1.1.0/cuda_installer.pyz --output cuda_installer.pyz

sudo python3 cuda_installer.pyz install_driver

My VM crashed the first time I ran the above commands. I had to restart the VM and run the commands again.

  1. Verify the driver is installed by running nvidia-smi

Install Blender

# Download Blender 4.2.3 LTS to the current directory
wget https://download.blender.org/release/Blender4.2/blender-4.2.3-linux-x64.tar.xz

# Extract blender to ./blender-4.2.3-linux-x64
sudo tar -xvf blender-4.2.3-linux-x64.tar.xz

Also install necessary blender dependencies:

# sudo apt-get install libglu1 libxi6 libgconf-2-4 libxkbcommon0 libxrender1 libsm6
sudo apt-get install libboost-all-dev libgl1-mesa-dev libglu1-mesa libsm-dev

Check blender is happy with:

./blender-4.2.3-linux-x64/blender --version

Housekeeping

Prepare our VM to receive our blend file

# Create projects directory and make it writable
mkdir projects
sudo chmod 777 projects
cd projects

Upload Blender file

Uploading via the ā€œFile Uploadā€ button works, but is very slow. Usually blend files are large so I suggest following these steps to first upload to Google Cloud Storage and then transfer to your VM.

  1. Install gsutil
  2. Run these commands, replacing paths and filenames as necessary:
# On your local machine:
gsutil cp my-blend-file.blend gs://my-bucket-name/my-subfolder

# On your GCP VM:
gsutil cp gs://my-bucket-name/my-subfolder/my-blend-file.blend .

Ensure blender uses GPU

  1. Create the following set_gpu.py script and upload it to your VM home directory:
import bpy
import re
scene = bpy.context.scene
scene.cycles.device = 'GPU'
prefs = bpy.context.preferences
prefs.addons['cycles'].preferences.get_devices()
cprefs = prefs.addons['cycles'].preferences
print(cprefs)
# Attempt to set GPU device types if available
for compute_device_type in ('CUDA', 'OPENCL', 'NONE'):
  try:
    cprefs.compute_device_type = compute_device_type
    print('Device found', compute_device_type)
    break
  except TypeError:
    pass
# Enable all CPU and GPU devices
for device in cprefs.devices:
    print('Considering', device)
  if not re.match('intel', device.name, re.I):
    print('Activating', device)
    device.use = True

Here are the shell commands that will do that for you:

cd ~
echo -e "import bpy\nimport re\nscene = bpy.context.scene\nscene.cycles.device = 'GPU'\nprefs = bpy.context.preferences\nprefs.addons['cycles'].preferences.get_devices()\ncprefs = prefs.addons['cycles'].preferences\nprint(cprefs)\nfor compute_device_type in ('CUDA', 'OPENCL', 'NONE'):\n  try:\n    cprefs.compute_device_type = compute_device_type\n    print('Device found', compute_device_type)\n    break\n  except TypeError:\n    pass\nfor device in cprefs.devices:\n  print('Considering', device)\n  if not re.match('intel', device.name, re.I):\n    print('Activating', device)\n    device.use = True" > set_gpu.py
chmod 777 set_gpu.py

blender-4.2.3-linux-x64/blender -b projects/my-blend-file.blend -P set_gpu.py
# blender-4.2.3-linux-x64/blender -b projects/bubblegum/render3.blend -P set_gpu.py

(optional) Configure render settings

If you need to correct some render settings in the blend file, but donā€™t want to reupload it, you can create a python script and run it against your blend file.

import bpy
scene = bpy.context.scene
scene.cycles.samples = 128
scene.cycles.adaptive_threshold = 0.01
scene.render.resolution_percentage = 100
scene.render.resolution_x = 1080
scene.render.resolution_y = 1920
scene.frame_start = 1
scene.frame_end = 10
scene.render.use_simplify = False
bpy.ops.wm.save_mainfile()
print("Render settings updated")

ch ~
echo -e "import bpy\nscene = bpy.context.scene\nscene.cycles.samples = 128\nscene.cycles.adaptive_threshold = 0.01\nscene.render.resolution_percentage = 100\nscene.render.resolution_x = 1080\nscene.render.resolution_y = 1920\nscene.frame_start = 1\nscene.frame_end = 330\nscene.render.use_simplify = False\nbpy.ops.wm.save_mainfile()\nprint('Render settings updated')" > set_render_settings.py

blender-4.2.3-linux-x64/blender -b projects/my-blend-file.blend -P set_render_settings.py

Relevant API docs

Render

Finally, we can render something.

Letā€™s render only the first 10 frames of our animation to test if everything is working.

# blender-4.2.3-linux-x64/blender -P set_gpu.py -b projects/bubblegum/render3.blend -o //out/_ -F OPEN_EXR -a -s 1 -e 10

Commandline args docs

Download render output

# On your GCP VM:
gsutil -m cp -r ./projects/my-project/out gs://my-bucket/my-project/out
# gsutil -m cp -r ./projects/bubblegum/out gs://ltm-blender-projects/bubblegum/out

# On your local machine:
gsutil -m cp -r gs://my-bucket-name/my-subfolder/out .
# gsutil -m cp -r gs://ltm-blender-projects/bubblegum/out .
ā† Home