To start, I tried every solution discussed in this thread and many others. I am running an XFX 9070 XT and a 7800X3D. The only thing that worked for me was GPU tuning after doing some logging. I used HWiNFO64 for logging with sensor polling set to 2000 ms and used ChatGPT to help review the logs and identify anomalies. So far this has been working without any DXGI crashes. I have logged two nights of BF6 RedSec sessions afterwards at roughly 2 to 3 hours each night and the logs no longer show the same anomalies that appeared before making these changes. On the second night I started re enabling things I had disabled while troubleshooting including HAGS, EXPO and PBO. The logs still came back clean.
The main thing that stood out in my original logs was aggressive GPU boost behavior.
Peak clock reached 3361 MHz.
Average clock was around 2859 MHz.
During gameplay the GPU was reaching 3300 plus MHz.
Example from the log:
3361 MHz core clock at 100 percent GPU utilization. That alone is not necessarily a problem but it became suspicious because it lined up with DXGI crashes. Another thing I noticed was unusual power behavior at max clocks.
Near the end of one session the GPU was reporting:
3361 MHz at 100 percent load.
GPU power dropped to roughly 102 to 107 watts TGP.
Earlier in the session I saw over 217 watts under similar load.
Normally you would expect high load plus high clock to equal high power.
Instead I had moments with high load, high clock and much lower power.
My theory ended up being:
BF6 DX12 workload to aggressive boost state to inconsistent workload behavior to DXGI device hang.
The tuning changes that helped were (Made these changes through AMD adrenalin):
Max Frequency Offset minus 150 MHz.
Voltage Offset minus 50 mV.
Power Limit minus 5 percent.
After applying these changes:
Clocks flattened out.
Logs became cleaner.
No DXGI crashes (knock on wood)
Later logs no longer showed the same behavior.
