2.76 48f7dd6 / GPU Experimental CUDA Out of Memory on Default Cube NVIDIA 780 Ti 3GB #46528

New Issue

H S · 2015-10-18T14:54:02+02:00

H S commented

2015-10-18 14:54:02 +02:00

System Information
Windows 10 Pro 64-Bit
i7 4930K
Nvidia GF 780 Ti:
2880 CUDA Cores
3GB 384-Bit GDDR5
Diver 3.5582
64 GB RAM

Blender Version:
Broken: 2.76 48f7dd6
Worked: 2.72, 2.71, 2.70, and 2.75 at least.

Short description of error
Rendering the default cube with Cycles experimental feature set using CUDA (Nvidia 780 Ti) results in an out of memory error. Normal feature set works.

Exact steps for others to reproduce the error
I opened Blender and set up the compute device to my GPU in User Preferences / System. Then I set the scene renderer to Cycles and changed the feature set to experimental. Hitting F12 results in the the render window opening and the render region tiles spawning as if rendering was going on, but no image appears. After the render finishes, I get the error in my console:

Read new prefs: C:\Users*******\AppData\Roaming\Blender Foundation\Blender\2.76\config\userpref.blend
AL lib: (EE) UpdateDeviceParams: Failed to set 44100hz, got 48000hz instead
found bundled python: C:\Program Files\Blender Foundation\Blender\2.76\python

SCRIPT STARTING

archimesh: added to phytonpath
archimesh: Imported multifiles

Sverchok loaded with 131 nodes
Have a nice day with sverchok

create isosurfer handlers...
isosurfer handler created successfully!
Loading MHX2 importer-runtime v 0.27
Loading MHX2 armature
build loaded
MHX2 armature loaded
shaders.py
shaders.py done
MHX2 successfully (re)loaded

CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)

Refer to the Cycles GPU rendering documentation for possible solutions:
http://www.blender.org/manual/render/cycles/gpu_rendering.html

CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)
CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)
CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)
CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)
Error: CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)

**System Information** Windows 10 Pro 64-Bit i7 4930K Nvidia GF 780 Ti: 2880 CUDA Cores 3GB 384-Bit GDDR5 Diver 3.5582 64 GB RAM **Blender Version:** Broken: 2.76 48f7dd6 Worked: 2.72, 2.71, 2.70, and 2.75 at least. **Short description of error** Rendering the default cube with Cycles experimental feature set using CUDA (Nvidia 780 Ti) results in an out of memory error. Normal feature set works. **Exact steps for others to reproduce the error** I opened Blender and set up the compute device to my GPU in User Preferences / System. Then I set the scene renderer to Cycles and changed the feature set to experimental. Hitting F12 results in the the render window opening and the render region tiles spawning as if rendering was going on, but no image appears. After the render finishes, I get the error in my console: Read new prefs: C:\Users\*******\AppData\Roaming\Blender Foundation\Blender\2.76\config\userpref.blend AL lib: (EE) UpdateDeviceParams: Failed to set 44100hz, got 48000hz instead found bundled python: C:\Program Files\Blender Foundation\Blender\2.76\python SCRIPT STARTING archimesh: added to phytonpath archimesh: Imported multifiles Sverchok loaded with 131 nodes Have a nice day with sverchok create isosurfer handlers... isosurfer handler created successfully! Loading MHX2 importer-runtime v 0.27 Loading MHX2 armature build loaded MHX2 armature loaded shaders.py shaders.py done MHX2 successfully (re)loaded CUDA error: Out of memory in cuMemAlloc(&device_pointer, size) Refer to the Cycles GPU rendering documentation for possible solutions: http://www.blender.org/manual/render/cycles/gpu_rendering.html CUDA error: Out of memory in cuMemAlloc(&device_pointer, size) CUDA error: Out of memory in cuMemAlloc(&device_pointer, size) CUDA error: Out of memory in cuMemAlloc(&device_pointer, size) CUDA error: Out of memory in cuMemAlloc(&device_pointer, size) Error: CUDA error: Out of memory in cuMemAlloc(&device_pointer, size)

H S commented

2015-10-18 14:54:02 +02:00

Changed status to: 'Open'

H S commented

2015-10-18 14:54:02 +02:00

Added subscriber: @91jnar91

Thomas Dinges commented

2015-10-18 15:10:57 +02:00

Added subscriber: @ThomasDinges

Thomas Dinges commented

2015-10-18 15:10:57 +02:00

Changed status from 'Open' to: 'Archived'

Thomas Dinges closed this issue

2015-10-18 15:10:57 +02:00

Thomas Dinges self-assigned this 2015-10-18 15:10:57 +02:00

Thomas Dinges commented

2015-10-18 15:10:57 +02:00

This is a known limitation, see http://www.blender.org/manual/render/cycles/features.html (the warning box).

We hope to improve this in the future, closing this as archived.

This is a known limitation, see http://www.blender.org/manual/render/cycles/features.html (the warning box). We hope to improve this in the future, closing this as archived.

H S commented

2015-10-18 15:24:06 +02:00

This comment was removed by @91jnar91

*This comment was removed by @91jnar91*

H S commented

2015-10-18 15:24:30 +02:00

In #46528#342248, @ThomasDinges wrote:
This is a known limitation, see http://www.blender.org/manual/render/cycles/features.html (the warning box).

We hope to improve this in the future, closing this as archived.

Has there been a modification to the requirements? Experimental feature set was fully functional on the same specs on 2.75

> In #46528#342248, @ThomasDinges wrote: > This is a known limitation, see http://www.blender.org/manual/render/cycles/features.html (the warning box). > > We hope to improve this in the future, closing this as archived. Has there been a modification to the requirements? Experimental feature set was fully functional on the same specs on 2.75

H S commented

2015-10-18 15:26:33 +02:00

Added subscriber: @candreacchio

H S commented

2015-10-18 15:26:33 +02:00

In #43200#280452, @candreacchio wrote:
Correct... The Cycles Experimental kernel takes up ~1200-1300mb of ram... your card only has 1280mb... you will not be able to render anything with the experimental kernel.

Based on info above, my graphics card should be able to handle the a simple cube on experimental. Also considering that fact that I have two 780 Tis totaling 6GB of GPU RAM

> In #43200#280452, @candreacchio wrote: > Correct... The Cycles Experimental kernel takes up ~1200-1300mb of ram... your card only has 1280mb... you will not be able to render anything with the experimental kernel. Based on info above, my graphics card should be able to handle the a simple cube on experimental. Also considering that fact that I have two 780 Tis totaling 6GB of GPU RAM

Thomas Dinges commented

2015-10-18 15:32:52 +02:00

It's called "Experimental" kernel for a reason. It might work, it might not work. Again there are plans to improve this (kernel split, runtime feature based compilation etc...) but that is a ToDo.

Brecht Van Lommel commented

2015-10-18 15:40:32 +02:00

Added subscriber: @brecht

Brecht Van Lommel commented

2015-10-18 15:40:32 +02:00

The memory that this kernel takes up is not a fixed number, it depends on the specific GPU architecture version, the number of CUDA cores, and new features added to Cycles.

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

2.76 48f7dd6 / GPU Experimental CUDA Out of Memory on Default Cube NVIDIA 780 Ti 3GB #46528

2.76 `48f7dd6` / GPU Experimental CUDA Out of Memory on Default Cube NVIDIA 780 Ti 3GB #46528