Inquiry Regarding Potential Software Issue in Performance Test v11.1.1002.0

[David Hsu]

Dear PassMark Support Team,

Our internal engineering team has been conducting tests using the PassMark Performance Test software and has encountered a potential issue related to the GPU testing module in the latest version. Below is a summary of our observations:

1. Testing on v10.2:
   • We performed multiple tests on the same machine using Performance Test v10.2.
   • No errors or abnormalities were observed during these tests.
2. Testing on v11.1.1002.0:
   • After switching to Performance Test v11.1.1002.0, the first few test runs were successful without any errors.
   • However, after further repeated testing on the same machine, we experienced the issue once.
   • This behavior did not occur at any point during our tests with v10.2.

Based on these findings, we suspect that the issue may be related to changes in the Performance Test software between versions 10.2 and 11, particularly in the GPU testing component. We would appreciate any guidance or further analysis on whether there have been known issues or changes that could explain the behavior.

Please let us know if there are additional diagnostics or logs we should provide to help with the investigation.

Thank you for your support.

[David (PassMark)]

Error was: Could not find a valid video adapter to use for the GPU compute tests.
CPU was: Celeron J6412
GPU was: Intel UHD Graphics

From the log file we see this
331.656s - GetCombinedAdapterList Getting OpenCL Adapters
331.719s - ERROR: clGetPlatformIDs failed (1) [-1001].

The function clGetPlatformIDs failed. Documentation is here.
https://registry.khronos.org/OpenCL/...atformIDs.html
The return code is CL_PLATFORM_NOT_FOUND_KHR (numeric value -1001)

This isn't a known issue.
If it initially worked, our guess is that the video card device driver crashed at some point during the tests. So maybe a bug in the driver.

There were many changes between V10 and V11. Full list is here.
https://www.passmark.com/products/pe...st/history.php

[David Hsu]

Dear David,

Thank you for your response. I would like to confirm that during our repeated tests, the issue occurred intermittently—sometimes it happened, and sometimes it did not. However, this issue never occurred in version V10. Are there any changes in version V11 that could potentially cause this issue?

The main concern is that all tests in version V10 ran without any problems, while in version V11, the issue occurs sporadically. I noticed in Build 1024 that there were related fixes and modifications to the 2D test. Could these changes possibly be the cause of the issue?

[David (PassMark)]

There were many changes between V10 and V11. Full list is here.
https://www.passmark.com/products/pe...st/history.php
We don't know which specific change provoked your driver crash (assuming that is what it is). Obviously the changes to the 2D, 3D & GPGPU tests would be the ones to look at.

Sometimes driver crashes write a log record into the Windows event log when the fail. But not always. So you could have a look at the Windows event log.

[David Hsu]

Dear David,

I would like to confirm if there is a possibility that the issue might be related to PassMark encountering errors when calling OpenGL or DirectX functions. As we discussed previously, discrepancies in test results seem to occur across different software versions. Therefore, we would like to check with you whether there might be any issues within PassMark's internal code when invoking OpenGL or DirectX functions.

Thank you for your support.

[David (PassMark)]

I understand it is tempting to shoot the messenger, and then ignore the real problem.

The PerformanceTest software has been run on millions of machines, without any other machine that we are aware of having this, CL_PLATFORM_NOT_FOUND_KHR error.
So it is more likely a problem with this particular machine, than a software bug (which would typically effect many machines).

Also the error is very clear about what & where the problem is. You have no hardware installed with OpenCL support when the error happened. So either the hardware is faulty, or the device driver for the hardware crashed. Of course driver issues are difficult problems to solve, so it is tempting to want to blame something else.