I have had the same experience with my XFX Qick RX6750 XT.
After a lot of testing and research i found the cause and the solution(s).
The cause is that the GPU Hotspot temp exceeds it's thermal limits thus shutdowns your system before it explodes, so thats actually good.
The solution i tried was to remove the backplate and the thermal pad that resides on the backside of the GPU.
I then took a copper plate and put new thermal pads at either side of the copper plate and placed it on the backside of the GPU. Then i re-assembled the backplate and tested it in various scenarios.
The result was very surprising.
The hotspot temp. and the gpu temp were much closer to each other (within 5 degrees celsius) and under full load it did almost hit the thermal threshold but luckily it stabilized at around 105 degrees celsius instead of going over the 115 degrees hard limit.
What was most suprising is that when i stopped the test, the hotspot temp. went down faster then the gpu temp !
So in my case i was able to succesfully solved the problem and even improved performance.
In short : you need to find a way to cool the backside of the GPU, in my case the backplate is plastic so i am thinking of cutting a whole into it and place a custom cooler on it. (if i am lucky i can use a AMD cpu cooler for that)
Or leave the backplate of entirely.
Anyhow, the issue never happened again so i am confident enough to say that the cause is the hotspot thermal limit exceedance so be inventive and try your own build perhaps. (be careful though ofcourse)
