Count me in soon. I still have an issue with the mem/timing/core tweaks. If it freezes the hardware, how and when can we be sure it worked ?
What I would do as a stress test is to run benchmarks with Repeats = 10 or even 20.
Statistically speaking, you need 100 trials for a 95% CI, but that would be too much stress on the hardware
and the results would be unfavorably skewed. If you are masochistic, you can turn on "Cool GPU down,"
which makes the test notoriously harder to pass.
On a more positive note, I just added extra parameters for memory timings to the upcoming version
for stability. We will see...