Passmark 11 crashes on Directx10 test on Intel Elkhart Lake

[ulfh]

I have run the Passmark PerformanceTest 11.0 Build 1009 evaluation version, using Windows 10 Pro 22H2, on one type of industrial 3.5" card equipped with Intel Elkhart Lake x6413E and x6414RE CPUs.

On both types, the DX10 test unfortunately stops working, but not right away, it first shows a black screen, like the GPU crashes and recovers once, and this always happens when a meteor is close by and directly on top of the floating island in the center of the picture.

It does recover from this, and continues rendering, but then a second black screen happens, perhaps a second crash, which it does not recover from, when a second meteor is on top of the island.

Then an error message is displayed: "The 3D test was stopped by the user before completing. The results for this test will not be used.". This error message is displayed behind the main screen as a second window, and a small black DX10 test windows is still there with it.

I can also mention that on a CPU card of the same type but with an Intel Elkhart Lake J6412, this does not happen, and the test works perfectly with the same Intel drivers.

However, if I run the test manually (Advanced) so that it is possible to vary the number of 3D objects, this fault only happens with a maximum number of both objects (50/250 islands and meteors), and with 8xMSAA activated at 1600x1200 or higher resolution.

As soon as I lower the MSAA to below 8, the number of objects below 50/250, or the resolution below 1600x1200, the benchmark works the way it should.

Also, all other 3D test work just like they should.

What could in your opinoin be causing this?

[David (PassMark)]

Maybe your system has extremely limited video RAM and you are exceeding it's low capabilities?

The hardware you are using doesn't sound like it was designed for 3D work (or any graphical work).

In some cases you can adjust some BIOS settings to use more of the main RAM as shared video RAM. So maybe check if that is an option for your machine.

Also check the video card device drivers are up to date.

[ulfh]

Thank you David for a quick answer.

You are correct, it is not really designed for 3D work, but it should work slowly, but at least not crash the benchmark.

I have checked the video RAM settings, and also maxed them all out in the BIOS, since it is an industrial card, there are quite detailed settings, but unfortunately it does not change the situation, it still crashes. I have also tried to increase total RAM to 16Gb, it does not help either.

The same card with a J6412 CPU, also using the same exact driver, does not crash, so there is a hardware difference.

I forgot to mention, there is also a popup that says "Application has been blocked from accessing Graphics hardware" to the lower right after it fails.

The system is using a shared RAM model, with an integrated Intel UHD Graphics, using the driver 31.0.101.2127, which Intel claims is the latest available, that was just released yesterday. Windows claims there is up to 3.8Gb of shared video memory available, so there doest not seem to be a lack of it, and PEtest.exe only uses about 0.7Gb of that for the DX10 test.

Strangely enough, I have now also tried an older version of the driver, 30.0.100.9978, and it does not crash, so this seems to be a problem with most newer Intel drivers, as I have tried several of the latest ones, but not that old, it is from 2022-09-08.

Also, this always happens when one meteor is really close in, filling almost the whole screen and going through the floating island, each and every time, which I think is very strange, it must be a combination of that one closeup object and the number of total objects, since if I lower the object numbers, it does not crash.

I have now also tested the old Futuremark Corporation 3DMark Vantage DX10 test suite, and it has several scenarios that use really complex graphics with thousands of 3D objects, and it works slowly but well at 1920x1080 and 8xMSAA. Sure, you only get 5-10 fps, but at least it works without crashing even on the latest driver.

We use PerformanceTest as a kind of reliability test that new hardware does not have any issues with new or older type graphics applications, you never know what industrial customers need, so it has become a dependable staple in our testing regimen of new hardware solutions.

I understand you probably cannot solve this problem, but would it be possible to send you a DEBUGMODE crash log or a DMP-file that was produced, so you could share any details that you can see there about any memory size issues?

[Richard (PassMark)]

I understand you probably cannot solve this problem, but would it be possible to send you a DEBUGMODE crash log or a DMP-file that was produced, so you could share any details that you can see there about any memory size issues?

Yes, you can send in the crash dump and debug log and we can take a look. If the crash is in PerformanceTest itself, we should be able to address it. If the crash is in the graphics drivers, then that is out of our hands.