Yeh but have you checked benchmark scores are actually rising?
Memory overclocking not what it used to be. You won't see snow or artifacts even at high clockspeeds but you aren't genuinely running that fast. It throttles to not make errors and likely you are either not running any faster or slightly less in real terms.
My VRAM taps out around +700Mhz so run +600Mhz
Thanks for pointing this out.
I can confirm I get performance increases all the way up to +1800 MHz, at which point Superposition crashes near the end of the benchmark.
+1750 produces the occassional geometry error (massive spikes in Unigine Heaven benchmark and games)
+1700 is therefore probably the sweet spot and what I will use for mining.
Some things to point out, in case others are interested:
1) Testing was done with the GPU core at fixed very low voltage / clocks (e.g. ~650 mV / 1100 MHz). This is to allow plenty of thermal and power headroom to truly see what the memory is capable of.
2) Performance increases are
tiny in something like Unigine Heaven (around 0.5% per 500 MHz).
3) For games, I plan to use the opposite setting - namely, max GPU clock and limiting the memory to "only" +1000 MHz. For me, the optimal curve seems to be +250 MHz at the low end (600 mV) and topping out around 1900 MHz at the high end (1000 mV). Generall, the GPU sits at 1650 MHz / 775 mV, where it's power-limited, but of course this varies on the game / scene being rendered.