CPU render will look different to GPU render and fails randomly #52625

Closed
opened 2017-09-02 16:27:35 +02:00 by Michael Klein · 21 comments

System Information
Win 10, GTX 880M, GTX 1080TI Founders Edition

Blender Version
Broken: blender-2.79.0-git.2ca1f29-windows64
Worked: na

Short description of error
Attached you'll finde a scene what has issues with GPU rendering. The scene is using the old shaders with a bevel shader (OSL).

The preview will not start or maybe it takes a very long time. Here nothing happened.
In GPU mode the preview starts as expected.

When I try to render with F12 in CPU mode it doesn't start or maybe it will start after a long time. Here nothing happened after 10 minutes.
When I render a small region with CPU it takes a long time till it starts to render.
When I start rendering in GPU mode it starts fast as expected ignoring OSL of course.

In the ZIP and attached you'll see a picture with a split screen. Left side is GPU. Right CPU. The results are different. Transparent parts with DOF will have extreme noise.

Exact steps for others to reproduce the error
Load the scene and press F12. Here CPU render will not start as expected.
Switch to GPU and press F12. Rendering will start.
Define a small region where the transparent parts are. Press F12. It will take really long but start. Result has noise at the transparent parts.
This might be a bug in the Cycles render engine or an issue with the shader tree.

170902_Blender_279_Cycles_CPU_Issue.zip
GPU_vs_GPU_279_Render.jpg

**System Information** Win 10, GTX 880M, GTX 1080TI Founders Edition **Blender Version** Broken: blender-2.79.0-git.2ca1f29-windows64 Worked: na **Short description of error** Attached you'll finde a scene what has issues with GPU rendering. The scene is using the old shaders with a bevel shader (OSL). The preview will not start or maybe it takes a very long time. Here nothing happened. In GPU mode the preview starts as expected. When I try to render with F12 in CPU mode it doesn't start or maybe it will start after a long time. Here nothing happened after 10 minutes. When I render a small region with CPU it takes a long time till it starts to render. When I start rendering in GPU mode it starts fast as expected ignoring OSL of course. In the ZIP and attached you'll see a picture with a split screen. Left side is GPU. Right CPU. The results are different. Transparent parts with DOF will have extreme noise. **Exact steps for others to reproduce the error** Load the scene and press F12. Here CPU render will not start as expected. Switch to GPU and press F12. Rendering will start. Define a small region where the transparent parts are. Press F12. It will take really long but start. Result has noise at the transparent parts. This might be a bug in the Cycles render engine or an issue with the shader tree. [170902_Blender_279_Cycles_CPU_Issue.zip](https://archive.blender.org/developer/F766811/170902_Blender_279_Cycles_CPU_Issue.zip) ![GPU_vs_GPU_279_Render.jpg](https://archive.blender.org/developer/F766838/GPU_vs_GPU_279_Render.jpg)
Author

Changed status to: 'Open'

Changed status to: 'Open'
Author

Added subscriber: @Renderbicks

Added subscriber: @Renderbicks
Member

Added subscriber: @LazyDodo

Added subscriber: @LazyDodo
Member

Even without the de-noiser on there quite a difference in the lighting of that tree between osl and svm.

svm.png

osl.png

The slow starting speed with osl has something to do with the environment map. i've noticed this before, updating the importance map with osl on takes forever

NOTE: This file won't reproduce the issue until you go to File->External data->Unpack all, in the packed state render will start virtually instantly.

Even without the de-noiser on there quite a difference in the lighting of that tree between osl and svm. ![svm.png](https://archive.blender.org/developer/F767067/svm.png) ![osl.png](https://archive.blender.org/developer/F767068/osl.png) The slow starting speed with osl has something to do with the environment map. i've noticed this before, updating the importance map with osl on takes *forever* NOTE: This file won't reproduce the issue until you go to File->External data->Unpack all, in the packed state render will start virtually instantly.

Added subscriber: @NahuelBelich

Added subscriber: @NahuelBelich

i have to be honest, not a developer im a user testing, downloaded the file started the render in cpu but didnt finish it i have to cancel, the render its huge and not sure where it is a transparent area. . .
That said.. . .
Tested on
I7 2600 16gb ram nvidia 1070 - Blender 2ca1f29
Render on cpu:
Start normal, didn't finish because of the long render time it would naturally a 2K image would take, so i canceled the render.
Render on Gpu:
Starts and renders and denoises, there is evident difference on the edges but couldn't notice anything particularly different.

Maybe a developer notice it right away or maybe a simpler smaller example its needed to test, not sure.

i have to be honest, not a developer im a user testing, downloaded the file started the render in cpu but didnt finish it i have to cancel, the render its huge and not sure where it is a transparent area. . . That said.. . . Tested on I7 2600 16gb ram nvidia 1070 - Blender 2ca1f29 Render on cpu: Start normal, didn't finish because of the long render time it would naturally a 2K image would take, so i canceled the render. Render on Gpu: Starts and renders and denoises, there is evident difference on the edges but couldn't notice anything particularly different. Maybe a developer notice it right away or maybe a simpler smaller example its needed to test, not sure.

In #52625#456556, @LazyDodo wrote:

The slow starting speed with osl has something to do with the environment map. i've noticed this before, updating the importance map with osl on takes forever

NOTE: This file won't reproduce the issue until you go to File->External data->Unpack all, in the packed state render will start virtually instantly.

Yep, unpacking the blend file makes slowwwww start, if starts at all

> In #52625#456556, @LazyDodo wrote: > The slow starting speed with osl has something to do with the environment map. i've noticed this before, updating the importance map with osl on takes *forever* > > NOTE: This file won't reproduce the issue until you go to File->External data->Unpack all, in the packed state render will start virtually instantly. Yep, unpacking the blend file makes slowwwww start, if starts at all
Author

In #52625#456566, @NahuelBelich wrote:
Render on cpu:
Start normal, didn't finish because of the long render time it would naturally a 2K image would take, so i canceled the render.
Render on Gpu:
Starts and renders and denoises, there is evident difference on the edges but couldn't notice anything particularly different.

Maybe a developer notice it right away or maybe a simpler smaller example its needed to test, not sure.

Thanks for looking into it. Yes, the packed file works with CPU. That's interesting.

With GPU you will have differences at the edges because GPU isn't supporting OSL what does here a bevel edge shader.

The main issue here is the slow-down/freeze of CPU rendering and the differences described in the above pictures. When you render the pcitures with GPU and CPU you will notice differences when comparing them.

> In #52625#456566, @NahuelBelich wrote: > Render on cpu: > Start normal, didn't finish because of the long render time it would naturally a 2K image would take, so i canceled the render. > Render on Gpu: > Starts and renders and denoises, there is evident difference on the edges but couldn't notice anything particularly different. > > Maybe a developer notice it right away or maybe a simpler smaller example its needed to test, not sure. Thanks for looking into it. Yes, the packed file works with CPU. That's interesting. With GPU you will have differences at the edges because GPU isn't supporting OSL what does here a bevel edge shader. The main issue here is the slow-down/freeze of CPU rendering and the differences described in the above pictures. When you render the pcitures with GPU and CPU you will notice differences when comparing them.
Author

CPU vs GPU.

Please note: GPU is not rendering the OSL bevels.
But there are differences in the background.

Minecraft Animal Farm with Detailed Scene Renderbricks_v03_CPU.jpg

Minecraft Animal Farm with Detailed Scene Renderbricks_v03_GPU.jpg

CPU vs GPU. Please note: GPU is not rendering the OSL bevels. But there are differences in the background. ![Minecraft Animal Farm with Detailed Scene Renderbricks_v03_CPU.jpg](https://archive.blender.org/developer/F767251/Minecraft_Animal_Farm_with_Detailed_Scene_Renderbricks_v03_CPU.jpg) ![Minecraft Animal Farm with Detailed Scene Renderbricks_v03_GPU.jpg](https://archive.blender.org/developer/F767250/Minecraft_Animal_Farm_with_Detailed_Scene_Renderbricks_v03_GPU.jpg)

Im just theorizing here, but could you setup a render side by side like the last one, but setting the bevel to 0 in OSL or directly disabling it.
Again im not a developer, but my instinct in production its telling me that the difference in value its because of the lac of high lights cause by the bevel that's missing in the gpu side that not support OSL.

Could be that or also my screen covers less colors than yours and im not noticing it, this last case happens to in a studio once, its not a crazy possibility

This last comment its not probably related to the long render times reported in the initial report its just to remove variables

Im just theorizing here, but could you setup a render side by side like the last one, but setting the bevel to 0 in OSL or directly disabling it. Again im not a developer, but my instinct in production its telling me that the difference in value its because of the lac of high lights cause by the bevel that's missing in the gpu side that not support OSL. Could be that or also my screen covers less colors than yours and im not noticing it, this last case happens to in a studio once, its not a crazy possibility This last comment its not probably related to the long render times reported in the initial report its just to remove variables
Member

Guys, i know you're trying to help, but this is a bug tracker, not a forum, lets try to keep the hypothesizing and chatter to a minimum. The issue is reproducible on our end. So for now just sit back and relax, nothing is required from your end at this point in time.

Guys, i know you're trying to help, but this is a bug tracker, not a forum, lets try to keep the hypothesizing and chatter to a minimum. The issue is reproducible on our end. So for now just sit back and relax, nothing is required from your end at this point in time.

Added subscriber: @brecht

Added subscriber: @brecht

From profiling this looks like a performance issue with the OpenImageIO texture cache and .hdr files, the workaround would be to save the background image texture as a .exr file.

The render differences look like they're just bevel differences, there can also be some differences in image texture filtering when rendering with OSL.

From profiling this looks like a performance issue with the OpenImageIO texture cache and .hdr files, the workaround would be to save the background image texture as a .exr file. The render differences look like they're just bevel differences, there can also be some differences in image texture filtering when rendering with OSL.
Author

In #52625#456607, @brecht wrote:
From profiling this looks like a performance issue with the OpenImageIO texture cache and .hdr files, the workaround would be to save the background image texture as a .exr file.

The render differences look like they're just bevel differences, there can also be some differences in image texture filtering when rendering with OSL.

Not only. Or the marked areas in my attached picture are caused by the bevels? You'll see (as shown in my original post) differences in the blurred orange and yellow part (CPU result is sharp grained) and some reflections in the brown parts. Maybe it's an effect caused by the OSL bevels.

170904_Blender_279_Cycles_CPU_Issue.jpg

> In #52625#456607, @brecht wrote: > From profiling this looks like a performance issue with the OpenImageIO texture cache and .hdr files, the workaround would be to save the background image texture as a .exr file. > > The render differences look like they're just bevel differences, there can also be some differences in image texture filtering when rendering with OSL. Not only. Or the marked areas in my attached picture are caused by the bevels? You'll see (as shown in my original post) differences in the blurred orange and yellow part (CPU result is sharp grained) and some reflections in the brown parts. Maybe it's an effect caused by the OSL bevels. ![170904_Blender_279_Cycles_CPU_Issue.jpg](https://archive.blender.org/developer/F768695/170904_Blender_279_Cycles_CPU_Issue.jpg)

You can remove the bevel node on those materials if you want to verify it. To me it looks like specular highlights from bevel blurred by depth of field, which can be quite noisy.

You can remove the bevel node on those materials if you want to verify it. To me it looks like specular highlights from bevel blurred by depth of field, which can be quite noisy.
Author

In #52625#456656, @brecht wrote:
You can remove the bevel node on those materials if you want to verify it. To me it looks like specular highlights from bevel blurred by depth of field, which can be quite noisy.

Affirmative. I turned off OSL and that was the reason. i will wait for your future built in "rounded edge" feature :-)

> In #52625#456656, @brecht wrote: > You can remove the bevel node on those materials if you want to verify it. To me it looks like specular highlights from bevel blurred by depth of field, which can be quite noisy. Affirmative. I turned off OSL and that was the reason. i will wait for your future built in "rounded edge" feature :-)
Author

This issue still exists. CPU and GPU create different results.

This issue still exists. CPU and GPU create different results.
Philipp Oeser removed the
Interest
Render & Cycles
label 2023-02-09 13:59:02 +01:00

The issue is difficult to reproduce anymore, since the OSL and SVM looks completely different in Blender 4.1 Beta. There is still a noticeable noise difference in the area of the tree though.

The issue is difficult to reproduce anymore, since the OSL and SVM looks completely different in Blender 4.1 Beta. There is still a noticeable noise difference in the area of the tree though.

I think the point here is to compare CPU and GPU with SVM, ignoring OSL. And from what I can tell they match now.

7 years ago there were some significant difference in the integrator between CPU and GPU, since cycles-x this is unified so I would expect noise to be similar.

I think the point here is to compare CPU and GPU with SVM, ignoring OSL. And from what I can tell they match now. 7 years ago there were some significant difference in the integrator between CPU and GPU, since cycles-x this is unified so I would expect noise to be similar.
Blender Bot added
Status
Archived
and removed
Status
Confirmed
labels 2024-03-07 15:05:58 +01:00
Brecht Van Lommel added
Status
Resolved
and removed
Status
Archived
labels 2024-03-07 15:06:03 +01:00
Author

Turning on OSL with GPU will cause a crash with Blender 4.1 Beta (24-03-07) when trying to render the attached scene.

Blender 3.6.9 reports an error:
Error: Requested OSL group data size (2480) is greater than the maximum supported with OptiX (2048)

Turning on OSL with GPU will cause a crash with Blender 4.1 Beta (24-03-07) when trying to render the attached scene. Blender 3.6.9 reports an error: _Error: Requested OSL group data size (2480) is greater than the maximum supported with OptiX (2048)_

Yes, we are tracking that in #118020.

Yes, we are tracking that in #118020.
Sign in to join this conversation.
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset Browser
Interest
Asset Browser Project Overview
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
EEVEE & Viewport
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
EEVEE & Viewport
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Priority
High
Priority
Low
Priority
Normal
Priority
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
5 Participants
Notifications
Due Date
The due date is invalid or out of range. Please use the format 'yyyy-mm-dd'.

No due date set.

Dependencies

No dependencies set.

Reference: blender/blender#52625
No description provided.