QA

QA Report: Resolving UE5 and Houdini Crashes on Dell Aurora R16

Symptoms:

When using Unreal Engine 5 (UE5) and Houdini on a Dell Aurora R16, the programs crash once the system memory usage exceeds 40%.

System Specifications:

• Processor: Intel(R) Core(TM) i9-14900KF @ 3.19 GHz

• Installed RAM: 32.0 GB (31.7 GB usable)

• System Type: 64-bit operating system, x64-based processor

Testing Procedure:

Benchmark tests were conducted using Houdini with the Axiom plugin. A total of 10 tests were performed and averaged to measure performance and identify issues before and after modifications.

Before Modifications:

• Axiom 3.2 - Quality Assurance - OpenCL - 5GB

• Houdini Version: 20.0.590 (Windows - x86_64)

• Device: Intel(R) Core(TM) i9-14900KF (1.1)

Performance Metrics:

• Total Time: 822.681 seconds

• Average Time Step: 8.227 seconds

• Minimum Time Step: 1.463 seconds

• Maximum Time Step: 13.012 seconds

• Average Performance: 3,900,105 voxels/s

• Minimum Performance: 3,723,173 voxels/s

• Maximum Performance: 4,017,740 voxels/s

• Average Active: 31,672,401 voxels

• Minimum Active: 5,779,456 voxels

• Maximum Active: 49,299,456 voxels

After Modifications:

• Axiom 3.2 - Quality Assurance - OpenCL - 5GB

• Houdini Version: 20.0.590 (Windows - x86_64)

• Device: Intel(R) Core(TM) i9-14900KF (1.1)

Performance Metrics:

• Total Time: 577.891 seconds

• Average Time Step: 5.779 seconds

• Minimum Time Step: 0.987 seconds

• Maximum Time Step: 9.214 seconds

• Average Performance: 5,629,395 voxels/s

• Minimum Performance: 5,268,067 voxels/s

• Maximum Performance: 6,011,014 voxels/s

• Average Active: 31,672,401 voxels

• Minimum Active: 5,779,456 voxels

• Maximum Active: 49,299,456 voxels

Solution Implemented:

1. Extended the TDR Delay:

   • The Timeout Detection and Recovery (TDR) feature in Windows was adjusted to allow more time for the GPU to complete its tasks. This can be modified via the Windows Registry.

2. Steps:

   • Open the Registry Editor (regedit).

   • Navigate to HKEY_LOCAL_MACHINE\System\CurrentControlSet\Control\GraphicsDrivers.

   • Add a new DWORD (32-bit) value named TdrDelay.

   • Set the value to 120 (seconds).

3. Disabled CPU SpeedStep and SpeedShift:

   • These features dynamically adjust the CPU's power consumption and performance. Disabling them can stabilize performance by preventing the CPU from downclocking under load.

4. Steps:

   • Enter the BIOS setup during boot (usually by pressing F2 or DEL).

   • Locate the settings for SpeedStep and SpeedShift.

   • Disable both features.

   • Save and exit the BIOS setup.

Effect of Modifications:

• The system now operates normally, albeit with a performance degradation of approximately 29%. This trade-off ensures stability and prevents crashes during high memory usage.

Probable Cause:

The crashes were likely due to the GPU timing out and the dynamic frequency scaling of the CPU. Extending the TDR delay provided the GPU with sufficient time to complete tasks without causing a timeout, while disabling SpeedStep and SpeedShift prevented the CPU from dynamically adjusting its frequency, which can cause instability under heavy load conditions.

By implementing these changes, the stability of UE5 and Houdini on the Dell Aurora R16 was significantly improved, allowing for more reliable operation during demanding tasks.