Immediate crash during cycles render directly after auto-save aborted with scene containing large texture #73230

New Issue

michael campbell · 2020-01-19T03:28:58+01:00

michael campbell commented

2020-01-19 03:28:58 +01:00

System Information
Operating system: Windows-10-10.0.17134-SP0 64 Bits
Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 431.86

Blender Version
Broken: version: 2.82 (sub 6), branch: master, commit date: 2020-01-06 20:54, hash: 3a35301fd1
Worked: (optional)

Short description of error
blender crashes with large 47mb jpg 27000 x 27000 image. Happens with both gpu and cpu rendering (so not vram related)

Exact steps for others to reproduce the error
load a big image into a material, apply to object and render. Can't provide image as it's nda, but the image uncompresses to 2.2 GB, total vram usage when rendering on GPU was 2.4 GB as the scene contains just one object with this material applied, lit with hdri which is just a 2k for lighting (transparent background).

Error log:

I0119 02:17:21.456306 12864 device_cuda.cpp:859] Buffer allocate: __tex_image_byte4_025, 42,511,660 bytes. (40.54M) in device memory
I0119 02:17:21.465284 12864 device_cpu.cpp:448] Texture allocate: __tex_image_byte4_025, 42,511,660 bytes. (40.54M)
Skipping auto-save, modal operator running, retrying in ten seconds...
aligned_malloc returns null: len=2976088740 in Cycles Alloc, total 1084153684
Error : Unrecognized Exception
Address : 0x00007FFB63A6A388
Module : C:\WINDOWS\System32\KERNELBASE.dll

**System Information** Operating system: Windows-10-10.0.17134-SP0 64 Bits Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 431.86 **Blender Version** Broken: version: 2.82 (sub 6), branch: master, commit date: 2020-01-06 20:54, hash: `3a35301fd1` Worked: (optional) **Short description of error** blender crashes with large 47mb jpg 27000 x 27000 image. Happens with both gpu and cpu rendering (so not vram related) **Exact steps for others to reproduce the error** load a big image into a material, apply to object and render. Can't provide image as it's nda, but the image uncompresses to 2.2 GB, total vram usage when rendering on GPU was 2.4 GB as the scene contains just one object with this material applied, lit with hdri which is just a 2k for lighting (transparent background). Error log: I0119 02:17:21.456306 12864 device_cuda.cpp:859] Buffer allocate: __tex_image_byte4_025, 42,511,660 bytes. (40.54M) in device memory I0119 02:17:21.465284 12864 device_cpu.cpp:448] Texture allocate: __tex_image_byte4_025, 42,511,660 bytes. (40.54M) Skipping auto-save, modal operator running, retrying in ten seconds... aligned_malloc returns null: len=2976088740 in Cycles Alloc, total 1084153684 Error : Unrecognized Exception Address : 0x00007FFB63A6A388 Module : C:\WINDOWS\System32\KERNELBASE.dll

michael campbell commented

2020-01-19 03:28:58 +01:00

Added subscriber: @3di

Stephen Swaney commented

2020-01-19 05:33:41 +01:00

Added subscriber: @StephenSwaney

Stephen Swaney commented

2020-01-19 05:33:41 +01:00

aligned_malloc returns null: len=2976088740

That means you ran out of contiguous memory. An attempt to allocate 2,976,088,740 bytes (roughly 3 GB) failed.

> aligned_malloc returns null: len=2976088740 That means you ran out of contiguous memory. An attempt to allocate 2,976,088,740 bytes (roughly 3 GB) failed.

Brecht Van Lommel commented

2020-01-19 12:02:48 +01:00

Added subscriber: @brecht

Brecht Van Lommel commented

2020-01-19 12:02:48 +01:00

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

Brecht Van Lommel commented

2020-01-19 12:02:48 +01:00

We require an example .blend and image file to reproduce the problem. If you can't share the original one, you can create an artificial one.

Normally I would expect allocating 3 GB of contiguous memory to work, if there is enough memory available. With virtual memory it does not actually need to be contiguous in RAM.

We require an example .blend and image file to reproduce the problem. If you can't share the original one, you can create an artificial one. Normally I would expect allocating 3 GB of contiguous memory to work, if there is enough memory available. With virtual memory it does not actually need to be contiguous in RAM.

michael campbell commented

2020-01-19 15:22:55 +01:00

Seems to be plenty of space left for the pagefile on the C drive, and the crash happens at around 26GB system ram usage, so there's enough space in system ram for the 3GB it's trying to allocate :

Edit: Just trying to recreate the crash in another file and it wont happen, so retried in the original file and it seems to be not crashing there either now.....strange, I haven't restarted the pc....although I did clear out the app data/temp folder. Would cycles have been pulling any data from that folder during render?

Seems to be plenty of space left for the pagefile on the C drive, and the crash happens at around 26GB system ram usage, so there's enough space in system ram for the 3GB it's trying to allocate : Edit: Just trying to recreate the crash in another file and it wont happen, so retried in the original file and it seems to be not crashing there either now.....strange, I haven't restarted the pc....although I did clear out the app data/temp folder. Would cycles have been pulling any data from that folder during render? ![image.png](https://archive.blender.org/developer/F8287423/image.png)

Brecht Van Lommel commented

2020-01-19 15:28:13 +01:00