Hi there,
I'm curious so I did some tests... with rather interesting results about Animation Rendering
Why this? Because...
... During rendering the GPU does not work @100% continuously.
For example, when switching to the next image.
The code not only seeks the GPU; it also uses the CPU, it reads and writes data on disk and in memory, needs time to execute commands, etc.
And during that time the GPUs are not used!
I thought about how to fill those blanks to optimize the use of GPU and so I thought to make two images simultaneously. When GPU not used for the first image, it will be for the second one! And vice-versa.
The secret: 2 Max instances!
Now, here are my tests and their results.
The Context:
TEST 1: All-in-one
Results:
Results:
Time to complete animation rendering:
TEST 3: 9min49s. Both GPUs on 2 instances is better for lite scenes. Better optimization for GPU usage; more fast but more vram usage too.
TEST 2: 10min19s. Separate GPU for each instances produced no real optimization for GPU usage but is a good compromise between rendering speed and vram usage.
TEST 1: 11min32s. Both GPUs on 1 instance stay the better option for High poly scene which needs more vram. "Low" GPU usage; more "slow" but more free vram.
Thanks for reading. May this help You.
...
NEXT TESTS: Static animation
I'm curious so I did some tests... with rather interesting results about Animation Rendering

Why this? Because...
... During rendering the GPU does not work @100% continuously.
For example, when switching to the next image.
The code not only seeks the GPU; it also uses the CPU, it reads and writes data on disk and in memory, needs time to execute commands, etc.
And during that time the GPUs are not used!
I thought about how to fill those blanks to optimize the use of GPU and so I thought to make two images simultaneously. When GPU not used for the first image, it will be for the second one! And vice-versa.
The secret: 2 Max instances!
Now, here are my tests and their results.
The Context:
- - target: rendering 20 frames from non-static animation (particleFlow: fallin bolt nuts)
- - resolution: 1920*1080
- - render passes: 100
- - light samples: 3
- - max depth: 16
- - Depth of field: Off
- - Glare: Off
- - Tone mapping: On
- - vignetting: 5.0
- - image blur: 1.0
TEST 1: All-in-one
3dsMax Instances: 1Results:
GPUs: 2
Frames: 20 (0 to 19)
Rendering Time: 11min 32sTEST 2: Split animation. Open same scene on a second Max instance and just change frames range (e.g. First half and Second half).
GPU #1 usage: 93%
GPU #1 vram usage: ~900
GPU #2 usage: 94%
GPU #2 vram usage: ~700
3dsMax Instances: 2> Both rendering started simultaneoustly
GPU on Max#1: GPU#1
Frames on Max#1: 10 (0 to 9)
GPU on Max#2: GPU#2
Frames on Max#2: 10 (10 to 19)
Results:
Rendering Time Max#1: 9min 58sTEST 3: Same as Test 2, but with both GPUs on both instances.
GPU #1 usage: 94%
GPU #1 vram usage: ~1100
Rendering Time Max#2: 10min 19s
GPU #2 usage: 95%
GPU #2 vram usage: ~900
3dsMax Instances: 2> Both rendering started simultaneoustly
GPU on Max#1: Both
Frames on Max#1: 10 (0 to 9)
GPU on Max#2: Both
Frames on Max#2: 10 (10 to 19)
Results:
Rendering Time Max#1: 9min 45sConclusions:
Rendering Time Max#2: 9min 49s
GPU #1 usage: 98%
GPU #1 vram usage: ~1200
GPU #2 usage: 99%
GPU #2 vram usage: ~1000
Time to complete animation rendering:
TEST 3: 9min49s. Both GPUs on 2 instances is better for lite scenes. Better optimization for GPU usage; more fast but more vram usage too.
TEST 2: 10min19s. Separate GPU for each instances produced no real optimization for GPU usage but is a good compromise between rendering speed and vram usage.
TEST 1: 11min32s. Both GPUs on 1 instance stay the better option for High poly scene which needs more vram. "Low" GPU usage; more "slow" but more free vram.
Thanks for reading. May this help You.
...
NEXT TESTS: Static animation
Comment