Regression: Image rendering fails while viewport rendering is in progress #106657
Operating system: OSX Ventura Version 13.3
Graphics card:AMD Radeon Pro Vega II Duo 32 GB
Broken: 3.5.0 Date 2023-03-29 02:56 Hash:
Worked: 3.4.1 Date 2022-12-19 17:00 Hash
Short description of error
CommandBuffer Failed: cycles_metal_integrator_compact_shadow_states
Exact steps for others to reproduce the error
- Load file provided
- Set ViewPortShading to MaterialPreview
- Then select ViewPortShading to Rendered
- While the viewport is rendering Select Render Image.
This will often cause the error. It may take an attempt or two but more likely to happen after at least one attempt.
In some case blender will throw the error in both the viewport and in the image window. In some cases...Blender will hang and require you to force quit. At other times it will crash the OS.
The larger the scene the more drastic the failure. In this simplified file it typically will recover.
I can't reproduce the error on an M1.
It seems to be something with AMD GPUs.
Operating system: macOS-13.2.1-arm64-arm-64bit 64 Bits
Graphics card: Metal API Apple M1 1.2
Can confirm. I experienced also crash when reproducing first time, but does not seem to be reproducible.
I can't reproduce this issue either on my Intel Mac
iMac (Retina 5K, 27-inch, 2020)
3.6 GHz 10-Core Intel Core i9
128 GB 2667 MHz DDR4
AMD Radeon Pro 5500 XT 8 GB
If there is other information you may need to track this down do not hesitate to ask. Currently Blender 3.5 crashes fairly regularly making it unusable. I have had to drop back to 3.4.1 for production work. If there are any log files or anything that you need ... happy to provide.
Can you test if it works better in the daily builds of 3.5.1 or 3.6.0?
There were various Metal related fixes. Not guaranteed to solve this particular issue, but still would be good to know.
Blender 3.5.1. Failed: cycles_metal_integrator_queued_paths_array
BlendblendCrash Log provided via file further down in thread
@Michael-Jones, is this something you can reproduce and investigate?
From the backtrace I would guess it has something to do with the background compilation added in 3.5.
3.6 alpha - more stable than 3.5.1 for this particular issue. Took a while to get it to crash.
Crash log in text file lower in thread.
Can you attach the logs as text files instead? It's a bit difficult to read the comments otherwise.
3.6 exhibits the same issues as 3.5.1 for me - I have had to revert to 3.4.1 as well
If I go to PREFEREFNCES => SYSTEM =>CYCLES RENDER DEVICES and uncheck (in my case) "Intel Core...." and leave ONLY the AMD PRocessor checked, it seems to work fairly well with what I have tested.
I went to Preferences > System > CYCLES RENDER DEVICES and unchecked in my case (Intel XEON W-3245 CPU # 3.20GHz) and crashed blender when I turned on the Render ViewPort. Log attached.
John - I am not one of the programmers, so I really don't know where to look in the logs for anthing - I am a user with similar issues to the ones you are having.
Hopefully one of the coders can dope out the issues.
just check in 3.6 Beta and unfortunately...still causing crashes.
I think may be a Vega specific issue as this exact issue doesn't seem to occur on the Navi cards. We're trying to narrow down a root cause.
I just tested in Blender Version 3.5.1 (3.5.1 2023-04-25)
Tried rendering and get the following errors:
CommandBuffer Failed: cycles_metal_prefix_sum
CommandBuffer Failed: cycles_metal_integrator_queued_shadow_paths_array
The good news is Blender is not crashing. It will now just skip the render and display the error or render the scene without the volumetric object.
Thank you for your continued help.
No due date set.
No dependencies set.
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?