Exploring realistic facial shapes with Blender – WIP.
Checkout out the showcase !
Reduced render time by 50% and removed banding on background by using combination of Blender’s 2.79 new features:
– preview rather than final sampling
– limited rather than full global illumination
– shadow catcher plate with composited background rather than usual studio floor setup, effectively not generating banding.
Blenderella, Advanced Character Modeling, originally released in 2011 as a training DVD for Blender Institute.
Check out the showcase!
We finally moved our rendering pipeline over to cloud providers.
Here was the render setup:
- 240 frames at 1920×1080 with cycles, split into 20 regions per frame.
- Total machines: 49
- Rendered files stored using Google Cloud Storage.
Render result summary:
- Total time: 5h 45m
- Average per frame: 1m 26s
- Average time per frame would be better if “aws” machines performed normally while using free tier instance. More details below.
VM instance provider summary:
bx – IBM Bluemix – 1 month trial ?!
- Only provided a Kubernetes cluster with 1 worker node in their free trial.
- Since we required access to many VM instances, integration to render pipeline was postponed for now.
aws – Amazon Web Services – free tier – EC2 instance 750 hours
- 15x t2.micro
- 1 core w/ 1Gb ram
- Intel(R) Xeon(R) CPU E5-2676 v3
- Average time per pixel = 12.46 ms
- @ 0:15 seconds, we can see the CPU taking 9-10 more time than all other providers.
az – Azure – 1 month trial – $250 CAD credit
- 16x Standard_F1s
- 1 core w/ 2Gb ram
- Intel(R) Xeon(R) CPU E5-2673 v3
- Average time per pixel = 1.76 ms
gcp – Google Cloud Platform – 1 year trial – $410.24 CAD credit ($300 USD)
- 16x Custom preemptive instance
- 1 core w/ 2Gb ram
- Intel(R) Xeon(R) CPU @ 2.30GHz
- Average time per pixel = 1.38 ms
vgt – Vagrant (free) / Virtualbox (free) running on Windows 10 (not free) – internal cluster
- 2x Virtualbox VM
- 4 HT cores w/ 3Gb ram
- Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
- Average time per pixel = 0.86 ms