RX6600 Troubleshooting: Fails at Windows but Passes on Linux

[dushiido]

Hello, I'd like to say my thanks for the wonderful tool you've made here, and personally it's been the farthest I've gotten to knowing about the nature of my problem. I'm still lost, however, so I'd appreciate the help.

Some weeks ago on Windows my games have started crashing (no overclocks; just defaults). This seems to happen regardless of gaming load, but it acts normally outside of it. I've updated drivers, did clean reinstalls with DDU, underclocked/volted and did 3DMark tests (which interestingly don't crash, doesn't overheat, and works like my GPU has no issue). I've also booted up and reset from a fresh Windows install-- problem persists. I updated with Win11 and it's still there unfortunately.

Everything seemed to point to a hardware problem until I moved to Linux, in which case my GPU passes the test and my games run perfectly. I'm not sure if it's hardware, so I need help interpretting my logs.

The following is my log after running it on Linux (a passed test).
memtest_vulkan (linux).log

And the following are three new Windows tests before I write this, they're all ran under the same conditions. Despite that, it can act differently sometimes on the terminal, like what you see on the second test.
memtest_vulkan1.log
memtest_vulkan2.log
memtest_vulkan3.log

Though I'm not sure if relevant, this usually happens on GPU-Z and MSi Afterburner monitoring when the error appears after ~4500 iterations. Notice how the graphs would have that gap when it happens before resuming:

Many thanks for reading, I'm still not sure why it works perfectly on Linux while failing on Windows all of a sudden, so I'd appreciate any interpretation.

[galkinvv]

In the second test it seems to had write error on the initial write, so all later checking treated this is a fail. So the error happened on initial performance raising.

That performane gap before test end is explicitely designed to test GPU reaction to power level changing, and it seems to detect error for your GPU.

So all seen error types seems to be related to power level changing. 3dmark tests it on a fixed high perf, so not noticing this. Do game craches happen near loading scenes when the performance typically drops?

Chances are that linux does power management a bit fidderent. Maybe it doesn't switch to lowest mode due to non-optimal desktop compositing implementation, or a bit different power management in the driver.

[dushiido]

Interesting! Appreciate the response. I suppose I've had a few run-ins where it crashes on transitions (like switching from the title screen to a character select), but it's not exclusively that; sometimes it's just on regular gaming where it suddenly stops.

Think I can be more confident where to take my troubleshooting now. Will try testing on a different PSU, grateful for the insight. Thanks for the tool.