GPU Crash Debugging and Reporting - o3de/o3de Wiki

This page is pending to move to the O3DE Documentation > Atom Renderer Guide. See PR here.

NOTE: a "device removal" in the logs constitutes a crash as far as GPU crashes go

So you've encountered a crash in the GPU 🐞, what are your options?

The most helpful thing you can do is file an issue, and provide as much information as possible. Standard reporting practices apply (minimal reproducible steps are awesome), but its also recommended that you try an alternative RHI (instructions below) and enable validation to see if additional information is produced that you could include in the bug report. For users with multiple GPUs (e.g. multiple discrete GPUs, or an integrated GPU), instructions on switching the GPU selected are described below as well.

Issue Reporting

When reporting a crash encountered, please indicate your GPU vendor (e.g. Nvidia, AMD, Intel, Apple, etc.), device name (e.g. RTX2080), and driver version. Next, indicate what RHI you were using (e.g. Vulkan, D3D12, Metal).

Testing alternative devices/RHIs

For machines that have multiple GPUs, you can inform the runtime to prioritize one device over another with a CLI option. At the command line, pass --forceAdapter="your_device". Here, your_device can be any case-insensitive partial match for the device you wish to select.

You can select a non-default RHI using the --rhi command line option. For example, --rhi=vulkan may be used to select the Vulkan RHI. Currently this only applies on Windows, where D3D12 is the default selected backend.

Enable Driver Validation

Pass one of the following values to the CLI option --rhi-device-validation= to enable RHI-level validation. For Vulkan, be sure that you have the SDK installed before enabling device validation (SDK link).

DRED/Aftermath

TODO: Some changes to the integration are planned, this will be documented when complete