Eevee samples + Motion blur steps = different behaviour compared to 2.90alpha #81576

New Issue

SecuoyaEx · 2020-10-09T13:58:56+02:00

SecuoyaEx commented

2020-10-09 13:58:56 +02:00

System Information
Operating system: Windows-10-10.0.18362-SP0 64 Bits
Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 436.15

Blender Version
Broken: version: 2.90.0, branch: master, commit date: 2020-08-31 11:26, hash: 0330d1af29
Worked: One of the 2.90 alpha versions, don't remember exactly which

Short description of error
Using Motion blur steps with Eevee samples is unintuitive and weird. For example, you get 16 final samples if you have 16 steps and 16 samples or less, but you get 272 final samples if you have 16 and 17 respectively

Longer description of error
Motion blur steps, other than making moving things look smoother, also behave as regular Eevee samples, jittering shadows, screen space reflections,, antialias, and so on, which is good. You can visually count the final samples by counting soft shadow jitters

Back in 2.90 alpha, Motion blur steps were simply multiplied over Eevee samples, they would work together like this:
2 steps, 2 samples = 4 final samples
2 steps, 3 samples = 6 final samples
3 steps, 2 samples = 6 final samples

But in 2.90 release, they seems to be working like this: If Motion blur steps ≥ Eevee samples, the final samples will only be as many as Motion blur steps, but if Motion blur steps < Eevee samples, they will multiply each other, so:

2 steps, 2 samples = 2 final samples
2 steps, 3 samples = 6 final samples
3 steps, 2 samples = 3 final samples

Is this a big deal? Well no, it's not the end of the world. But it's unintuitive. Gets exponentially worse the more motion blur steps you want to use, but you could argue that if you use more motion blur steps, you don't need to suply extra samples anymore to converge things like shadows, and that the post processing component is good enough to smooth the inbetweens. However it can be a tough thing to balance with memory and rendertimes, and it's a shame not to have that range of possible numbers available.

Exact steps for others to reproduce the error
Create a new scene with a cube, one light and a plane, set Max Blur to 0. Count the resulting number of shadow jitters when rendering and try these different values, then compare with 2.90 alpha.

Same amount of steps and samples,
More steps than samples,
More samples than steps

**System Information** Operating system: Windows-10-10.0.18362-SP0 64 Bits Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 436.15 **Blender Version** Broken: version: 2.90.0, branch: master, commit date: 2020-08-31 11:26, hash: `0330d1af29` Worked: One of the 2.90 alpha versions, don't remember exactly which **Short description of error** Using Motion blur steps with Eevee samples is unintuitive and weird. For example, you get **16** final samples if you have 16 steps and 16 samples or less, but you get **272** final samples if you have 16 and 17 respectively **Longer description of error** Motion blur steps, other than making moving things look smoother, also behave as regular Eevee samples, jittering shadows, screen space reflections,, antialias, and so on, which is good. You can visually count the final samples by counting soft shadow jitters Back in 2.90 alpha, Motion blur steps were simply multiplied over Eevee samples, they would work together like this: 2 steps, 2 samples = 4 final samples 2 steps, 3 samples = 6 final samples 3 steps, 2 samples = 6 final samples But in 2.90 release, they seems to be working like this: If Motion blur steps ≥ Eevee samples, the final samples will only be as many as Motion blur steps, but if Motion blur steps < Eevee samples, they will multiply each other, so: 2 steps, 2 samples = 2 final samples 2 steps, 3 samples = 6 final samples 3 steps, 2 samples = 3 final samples Is this a big deal? Well no, it's not the end of the world. But it's unintuitive. Gets exponentially worse the more motion blur steps you want to use, but you could argue that if you use more motion blur steps, you don't need to suply extra samples anymore to converge things like shadows, and that the post processing component is good enough to smooth the inbetweens. However it can be a tough thing to balance with memory and rendertimes, and it's a shame not to have that range of possible numbers available. **Exact steps for others to reproduce the error** Create a new scene with a cube, one light and a plane, set Max Blur to 0. Count the resulting number of shadow jitters when rendering and try these different values, then compare with 2.90 alpha. Same amount of steps and samples, More steps than samples, More samples than steps

SecuoyaEx commented

2020-10-09 13:58:56 +02:00

Added subscriber: @SecuoyaEx

Richard Antalik commented

2020-10-13 15:43:02 +02:00

Added subscriber: @iss

Richard Antalik commented

2020-10-13 15:43:02 +02:00

Changed status from 'Needs Triage' to: 'Archived'

Richard Antalik closed this issue

2020-10-13 15:43:02 +02:00

Richard Antalik self-assigned this 2020-10-13 15:43:02 +02:00

Richard Antalik commented

2020-10-13 15:43:02 +02:00

I am testing with 2.90.1 so there may be some changes, but I don't think there are.

This seems to be documented in https://docs.blender.org/manual/en/2.90/render/eevee/render_settings/motion_blur.html#bpy-types-sceneeevee-motion-blur

I have confirmed documented behavior in my setup to be true.

2 steps, 3 samples = 6 final samples

Seems to be not true, it does 4 samples not 6. Same goes for 16 steps with 16 samples - it only does 16 samples

I think it would be best to find better way to count samples than counting shadow jitter here.

I see no bug here so I will close this report. If you think you can disprove my claims, please upload sample .blend file and method how to test.

I am testing with 2.90.1 so there may be some changes, but I don't think there are. This seems to be documented in https://docs.blender.org/manual/en/2.90/render/eevee/render_settings/motion_blur.html#bpy-types-sceneeevee-motion-blur I have confirmed documented behavior in my setup to be true. > 2 steps, 3 samples = 6 final samples Seems to be not true, it does 4 samples not 6. Same goes for 16 steps with 16 samples - it only does 16 samples I think it would be best to find better way to count samples than counting shadow jitter here. I see no bug here so I will close this report. If you think you can disprove my claims, please upload sample .blend file and method how to test.

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Eevee samples + Motion blur steps = different behaviour compared to 2.90alpha #81576