Cycles baking: "CUDA error: Launch exceeded timeout in ..." #40370

Closed
opened 2014-05-26 21:54:43 +02:00 by Wojciech Matyjewicz · 15 comments

System Information and Blender Version

  • GPU: NVIDIA GTX 670
  • The error occurs both on Linux and Windows
    • Linux: Blender build b33d83bf51, Debian jessie x64, nvidia-glx 331.67-2, nvidia-cuda-toolkit 5.5.22-3 (6.0 not yet available in jessie)
    • Windows: Blender 2.71, Windows 8 64-bit, NVIDIA driver version 337.88 (the most recent ones)

Short description of error
Baking COMBINED pass with Cycles on a GPU does not work for higher number of samples. It fails with a series of errors "CUDA error: Launch exceeded timeout in ...". See log.txt.

Exact steps for others to reproduce the error

  • Open cuda-error.blend. It should have number of samples set to 32.
  • Click 'Bake'. No error should occur.
  • Change the number of samples to 64.
  • Click 'Bake'. It should fail with errors described above.
**System Information and Blender Version** - GPU: NVIDIA GTX 670 - The error occurs both on Linux and Windows - Linux: Blender build b33d83bf51, Debian jessie x64, nvidia-glx 331.67-2, nvidia-cuda-toolkit 5.5.22-3 (6.0 not yet available in jessie) - Windows: Blender 2.71, Windows 8 64-bit, NVIDIA driver version 337.88 (the most recent ones) **Short description of error** Baking COMBINED pass with Cycles on a GPU does not work for higher number of samples. It fails with a series of errors "CUDA error: Launch exceeded timeout in ...". See [log.txt](https://archive.blender.org/developer/F91242/log.txt). **Exact steps for others to reproduce the error** - Open [cuda-error.blend](https://archive.blender.org/developer/F91243/cuda-error.blend). It should have number of samples set to 32. - Click 'Bake'. No error should occur. - Change the number of samples to 64. - Click 'Bake'. It should fail with errors described above.

Changed status to: 'Open'

Changed status to: 'Open'

Added subscriber: @wmatyjewicz

Added subscriber: @wmatyjewicz

Added subscribers: @brecht, @dfelinto

Added subscribers: @brecht, @dfelinto

@brecht could you confirm this? no CUDA here

@brecht could you confirm this? no CUDA here

Yes, I can confirm it. For graphics cards used for display the operating system puts a limit on how long a kernel can run, and if that is exceeded it is cancelled.

The only way to solve this is to split up the baking code so it only computes one sample each time the kernel is called, the same way regular rendering only does one sample at a time. That's quite a bit of work though, and it's a bit late in the release cycle. It may be best to disable CUDA baking for the release.

Yes, I can confirm it. For graphics cards used for display the operating system puts a limit on how long a kernel can run, and if that is exceeded it is cancelled. The only way to solve this is to split up the baking code so it only computes one sample each time the kernel is called, the same way regular rendering only does one sample at a time. That's quite a bit of work though, and it's a bit late in the release cycle. It may be best to disable CUDA baking for the release.
Member

Added subscriber: @MartijnBerger

Added subscriber: @MartijnBerger
Member

Is it possible to bake using multiple kernel invocations? As the timeout is per kernel invocation this could provide a way around this limitation.

Is it possible to bake using multiple kernel invocations? As the timeout is per kernel invocation this could provide a way around this limitation.

Yes, that's what should be done. It's already split now by the number of pixels, but if you have many samples a single pixel can also take long, so it also needs to be split by the number of samples.

Yes, that's what should be done. It's already split now by the number of pixels, but if you have many samples a single pixel can also take long, so it also needs to be split by the number of samples.

I can confirm that as a workaround one can configure BIOS and OS not to use the CUDA device as the display device. At least for Linux. I have not managed to configure Windows not to treat the NVIDIA GPU as a potential display device (even though no display was connected to it).

I can confirm that as a workaround one can configure BIOS and OS not to use the CUDA device as the display device. At least for Linux. I have not managed to configure Windows not to treat the NVIDIA GPU as a *potential* display device (even though no display was connected to it).

This issue was referenced by e4e58d4612

This issue was referenced by e4e58d46128dc7fe4fb9b881d73b38173f00f5c3

Changed status from 'Open' to: 'Resolved'

Changed status from 'Open' to: 'Resolved'

Closed by commit e4e58d4612.

Closed by commit e4e58d4612.
istavang commented 2014-08-23 18:59:53 +02:00 (Migrated from localhost:3001)

Added subscriber: @istavang

Added subscriber: @istavang
istavang commented 2014-08-23 18:59:53 +02:00 (Migrated from localhost:3001)

is this resolved ? I still get cuda error on gtx670 driver ver 340 52
win 8.1 64b

erroe msg: not found cuModule getglobal cumem and cubytes cumodulename when set to rendered mode

is this resolved ? I still get cuda error on gtx670 driver ver 340 52 win 8.1 64b erroe msg: not found cuModule getglobal cumem and cubytes cumodulename when set to rendered mode

@istavang this report is related to baking. If you get an error when setting to rendered mode then it's not a baking bug. Please create a new but report for that.

@istavang this report is related to baking. If you get an error when setting to rendered mode then it's not a baking bug. Please create a new but report for that.
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#40370
No description provided.