The most common problem we’ve had with long-term overclocking on Intel’s 45 nm process isn’t dead parts, but damaged parts that still appear to function correctly when set to far lower speeds. Intel wasn’t joking when it told us two years ago that 1.45V was the absolute limit for then-current parts, as we’ve seen everything from Core 2 to Core i7 devices damaged over a period of a few weeks using only a few hundredths of a volt more.
The bad news is that our current Core i7-930 wouldn’t exceed 4.1 GHz at anything less than 1.45V, even though our previous SBM machine ran at 4.3 GHz using only 1.36V with a lower-model processor of the same core revision. Receiving a highly overclockable processor usually happens by chance, and this time we were not so lucky.
The good news is that heat was never a problem. Most of our Nehalem-based processors will run with stability right up to the point at which they reset (between 100 degrees and 101 degrees Celsius), and ours didn’t get anywhere near that level before it simply ran out of overclocking capability. Using a faster fan that dropped the CPU core temperature by five degrees Celsius offered no improvement in its overclocking headroom. Higher voltage helped, but when we saw the core voltage fluctuate above 1.45V, we decided to back off for the sake of longevity.
Next, we added Nvidia Control Panel software, which enables the overclocking tab in its driver controls. Starting at a lowly 607 MHz GPU and GDDR5-3482, these cards were initially stable at an incredible 701 MHz GPU and GDDR5-3896. After an hour or more, the cards became unstable, a problem resolved by dropping the graphics memory clock. Another 30 minutes passed before our new overclock became unstable, a problem resolved by dropping the GPU clock. This type of progressive instability continued until we finally got through a complete benchmark set and stability test at 643 MHz GPU, GDDR5-3584.
Whether the overclocking capabilities of these cards continues to decline is a question only fate can answer, since we ran out of time after two days of additional tests. Full-load temperatures were held constant throughout all test conditions with the GeForce GTX 470 automatic fan controls and we never reached the cooling limits of either card at any frequencies.