Limited To Graphics Overclocking
Our hands are largely tied when it comes to overclocking, yet again. Base clock settings are out, as is the option to alter our processor's multiplier ratio. Intel even keeps us limited to DDR3-1333 data rates, compelling us to push lower latencies in the search for better performance.
Unfortunately, we lose user-selectable DIMM voltage with Gigabyte's B75M-D3V, and the only way to exceed 1.5 V is through an XMP memory profile. Reducing timings one at a time, we found stability at 8-9-8-22 1T through three passes of Memtest86+ and an hour of Prime95’s blend test. According to SiSoftware's Sandra diagnostic, however, this actually penalized us with a loss of bandwidth. Instead of messing with sub-timings to eke out insignificant gains, we settled for stock 9-9-9-24 latencies and shifted focus to graphics tuning.
Comparing one GeForce card to another gave us the perfect opportunity to embark on more aggressive GPU overclocking techniques than what we've used in the past. After dialing in the highest stable overclock at a stock 0.975 V, we made incremental bumps in voltage using MSI's Afterburner utility, increasing core clock rate and testing stability along the way.
After every 0.025 V bump, the GPU rewarded us with an extra 25 MHz or so, eventually settling on a 960 MHz limit at 1.050 V. MSI’s cooling solution remained super effective; lengthy stability testing using automatic fan settings saw our GPU top out at 70 degrees, triggering a mere 53% fan duty cycle. The way we were scaling, it might have been possible to hit 980-990 MHz at the maximum 1.087 V setting. However, even with cooling to spare, I didn’t want to run SBM benchmarks at voltages that exceeded my own comfort zone for day-to-day use.
We didn't bother trying to push the GDDR5 memory beyond its stable 1125 MHz (4500 MT/s) setting, and we ended up testing at a final combination of 950 and 1102.5 MHz for the core and memory, respectively.
