CPU render + OptiX denoiser leaves empty tiles unprocessed (Blender from Master) #84650

Closed
opened 2021-01-12 18:42:39 +01:00 by YAFU · 25 comments

System Information
Operating system: Linux-5.4.0-14-generic-x86_64-with-debian-bullseye-sid 64 Bits
Graphics card: GeForce GTX 960/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.32.03

CPU: i7 3770

Blender Version
Broken: version:

  • 2.92.0 Beta, branch: master, commit date: 2021-01-17 10:36, hash: 2ee2f87f29
  • 2.92.0 Alpha, branch: master, commit date: 2021-01-11 21:30, hash: acbda123ad
    Worked:
  • 2.92.0 Alpha, branch: master, commit date: 2021-01-07 22:28, hash: 61f1faac3f (for @rjg)
  • 2.91.0

Short description of error
When I render in Cycles with Sampling> Denoising> Render> Optix, strange things happen. The resulting render is left with empty tiles / holes and if you cancel (Esc) the render job you get the following error message:

CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816)

CPU_OptiX denoiser problem.blend
cpu_optix_denoiser.jpg

Depending on the scene there may be more than one empty tile.

Probably related to this fix?:
https://developer.blender.org/T83925#1087610

**System Information** Operating system: Linux-5.4.0-14-generic-x86_64-with-debian-bullseye-sid 64 Bits Graphics card: GeForce GTX 960/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.32.03 CPU: i7 3770 **Blender Version** Broken: version: - 2.92.0 Beta, branch: master, commit date: 2021-01-17 10:36, hash: `2ee2f87f29` - 2.92.0 Alpha, branch: master, commit date: 2021-01-11 21:30, hash: `acbda123ad` Worked: - 2.92.0 Alpha, branch: master, commit date: 2021-01-07 22:28, hash: `61f1faac3f` (for @rjg) - 2.91.0 **Short description of error** When I render in Cycles with Sampling> Denoising> Render> Optix, strange things happen. The resulting render is left with empty tiles / holes and if you cancel (Esc) the render job you get the following error message: ``` CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816) ``` [CPU_OptiX denoiser problem.blend](https://archive.blender.org/developer/F9561290/CPU_OptiX_denoiser_problem.blend) ![cpu_optix_denoiser.jpg](https://archive.blender.org/developer/F9561291/cpu_optix_denoiser.jpg) Depending on the scene there may be more than one empty tile. Probably related to this fix?: https://developer.blender.org/T83925#1087610
Author

Added subscriber: @YAFU

Added subscriber: @YAFU

#84793 was marked as duplicate of this issue

#84793 was marked as duplicate of this issue
YAFU changed title from CPU render + OptiX denoiser leaves empty tiles unprocessed to CPU render + OptiX denoiser leaves empty tiles unprocessed (Blender from Master) 2021-01-12 18:43:13 +01:00

Added subscriber: @rjg

Added subscriber: @rjg

Changed status from 'Needs Triage' to: 'Needs User Info'

Changed status from 'Needs Triage' to: 'Needs User Info'

I couldn't reproduce the issue on Windows, it renders fine and I'm not seeing that error.

  • Have you checked if this is the first error message you're getting?
  • Have you checked that your TDR delay is high enough?
  • What is the most recent version that worked correctly on your system?
I couldn't reproduce the issue on Windows, it renders fine and I'm not seeing that error. - Have you checked if this is the first error message you're getting? - ~~Have you checked that your TDR delay is high enough?~~ - What is the most recent version that worked correctly on your system?
Author

The latest version I know of that works without problems is 2.91.0. From master in versions prior to January 5 the file Crash Blender, this before this Fix:
https://developer.blender.org/rB3373d14b1b05b2ee5dc88efff5dc8b1a5fe20f36

TDR delay is primarily a Windows problem I understand.
The aforementioned error message is the only one I get when finishing the render or canceling the render. After that, without closing blender the next time I try to render I get the message: Failed to create CUDA context (Illegal address)

The latest version I know of that works without problems is 2.91.0. From master in versions prior to January 5 the file Crash Blender, this before this Fix: https://developer.blender.org/rB3373d14b1b05b2ee5dc88efff5dc8b1a5fe20f36 TDR delay is primarily a Windows problem I understand. The aforementioned error message is the only one I get when finishing the render or canceling the render. After that, without closing blender the next time I try to render I get the message: Failed to create CUDA context (Illegal address)

Changed status from 'Needs User Info' to: 'Needs Triage'

Changed status from 'Needs User Info' to: 'Needs Triage'

Changed status from 'Needs Triage' to: 'Needs User Info'

Changed status from 'Needs Triage' to: 'Needs User Info'

Unfortunately, I don't have a Nvidia GPU under Linux, so I can't test this specific case. You're right about TDR that is Windows only.

Could you please run Blender with the following debug flags and upload the log file.

./blender --debug --debug-gpu > ~/blender_debug_output.txt 2>&1
Unfortunately, I don't have a Nvidia GPU under Linux, so I can't test this specific case. You're right about TDR that is Windows only. Could you please run Blender with the following debug flags and upload the log file. ```lines ./blender --debug --debug-gpu > ~/blender_debug_output.txt 2>&1 ```
Author

Ok I'll do it. But first I would like to know why when I open Blender in the terminal or in the log, confidential information such as my email address, user and username appears. I am opening blender without being logged into any blender site, and I have also downloaded blender without being logged into any blender site. How does Blender collect that information? The information shown seems to be from my old gmail account that I was registered with here, so I'm really confused.

I guess it could be some addon?

EDIT:
Ok, I have run with an empty config folder and the confidential information does not appear. I am going to investigate which addon it is. BlenderKit is the only one that I remember having to sign up for.
Here is the log file:
blender_debug_output.txt

Ok I'll do it. But first I would like to know why when I open Blender in the terminal or in the log, confidential information such as my email address, user and username appears. I am opening blender without being logged into any blender site, and I have also downloaded blender without being logged into any blender site. How does Blender collect that information? The information shown seems to be from my old gmail account that I was registered with here, so I'm really confused. I guess it could be some addon? EDIT: Ok, I have run with an empty config folder and the confidential information does not appear. I am going to investigate which addon it is. BlenderKit is the only one that I remember having to sign up for. Here is the log file: [blender_debug_output.txt](https://archive.blender.org/developer/F9577312/blender_debug_output.txt)

Blender doesn't collect, save or require an email address or username. Perhaps that is printed by some add-on the requires an online account?

The log file may contain paths and therefore the name of your current active user on your system. That isn't considered a secret though, but you can redact that information from the log file if you want.

Blender doesn't collect, save or require an email address or username. Perhaps that is printed by some add-on the requires an online account? The log file may contain paths and therefore the name of your current active user on your system. That isn't considered a secret though, but you can redact that information from the log file if you want.
Author

Thank you. I have edited the previous message.

Thank you. I have edited the previous message.
Author

If from Debug menu I set "CUDA Streams = 10" the number of empty tiles changes some times:
CUDA streams 10.jpg

CUDA streams 10_2.jpg

If from Debug menu I set "CUDA Streams = 10" the number of empty tiles changes some times: ![CUDA streams 10.jpg](https://archive.blender.org/developer/F9577569/CUDA_streams_10.jpg) ![CUDA streams 10_2.jpg](https://archive.blender.org/developer/F9577578/CUDA_streams_10_2.jpg)
Author

Added subscriber: @pmoursnv

Added subscriber: @pmoursnv
Author

Hi @pmoursnv , I dare to quote you since there are already several reports similar to this and because 2.92 is beta now:

https://developer.blender.org/T84711

https://developer.blender.org/T84746

Thanks.

Hi @pmoursnv , I dare to quote you since there are already several reports similar to this and because 2.92 is beta now: https://developer.blender.org/T84711 https://developer.blender.org/T84746 Thanks.

Added subscriber: @zrp

Added subscriber: @zrp

I'm getting the same error on the above file. and it does occur in large larger scale environments too.

Error: CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816)CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816)

System: Windows 10 Pro 64bit, Ryzen 5800x, Nvidia 3070 461.09

Blender Version 2.92 Beta

error_cuda.PNG

I'm getting the same error on the above file. and it does occur in large larger scale environments too. Error: CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816)CUDA_ERROR_ILLEGAL_ADDRESS in cuStreamSynchronize(cuda_stream[thread_index]) (device_optix.cpp:816) System: Windows 10 Pro 64bit, Ryzen 5800x, Nvidia 3070 461.09 Blender Version 2.92 Beta ![error_cuda.PNG](https://archive.blender.org/developer/F9580793/error_cuda.PNG)

I'm able to reproduce this in 2.92.0 Beta, branch: master, commit date: 2021-01-17 10:36, hash: 2ee2f87f29 with GeForce GTX 1080 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.89. Perhaps the last time I tested it I got lucky and it didn't happen, will test older versions now.

This does seem to be a genuine issue, definitely not caused by TDR delay (rendering takes less time than the TDR) or the other typical reasons.

I'm able to reproduce this in 2.92.0 Beta, branch: master, commit date: 2021-01-17 10:36, hash: `2ee2f87f29` with GeForce GTX 1080 Ti/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 460.89. Perhaps the last time I tested it I got lucky and it didn't happen, will test older versions now. This does seem to be a genuine issue, definitely not caused by TDR delay (rendering takes less time than the TDR) or the other typical reasons.

Changed status from 'Needs User Info' to: 'Confirmed'

Changed status from 'Needs User Info' to: 'Confirmed'

I'm confirming this as it appears to be a bug in OptiX denoiser implementation. The issue is very reliably reproducible for me in the referenced build.

@pmoursnv @brecht This is likely something for you to look at.

I'm confirming this as it appears to be a bug in OptiX denoiser implementation. The issue is very reliably reproducible for me in the referenced build. @pmoursnv @brecht This is likely something for you to look at.

Added subscriber: @brecht

Added subscriber: @brecht
Member

This is a bug introduced with the tile stealing implementation (517ff40b12). Since the OptiX denoiser runs on the GPU, the tile stealing code erroneously steals CPU tiles and moves them to the OptiX device. But in this configuration the OptiX device was only set up for denoising, not rendering, so it crashes. Will look for a fix.

This is a bug introduced with the tile stealing implementation (517ff40b12). Since the OptiX denoiser runs on the GPU, the tile stealing code erroneously steals CPU tiles and moves them to the OptiX device. But in this configuration the OptiX device was only set up for denoising, not rendering, so it crashes. Will look for a fix.

Added subscriber: @Aiduss

Added subscriber: @Aiduss

This issue was referenced by 4fde594fda

This issue was referenced by 4fde594fda13abf98437bc5d0012decc2bd3d5f6
Member

Changed status from 'Confirmed' to: 'Resolved'

Changed status from 'Confirmed' to: 'Resolved'
Patrick Mours self-assigned this 2021-01-18 15:33:16 +01:00
Thomas Dinges added this to the 2.92 milestone 2023-02-08 16:14:53 +01:00
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#84650
No description provided.