The Perils Of Overclocking
This editor has been using 1.45 V to overclock ever since Intel’s 45nm Wolfdale cores were introduced. So far, we haven’t incurred any CPU damage as the result of voltage alone, though a certain Core i7 motherboard failed repeatedly from high VRM stress and eventually killed a CPU. But these are the new, more efficient motherboards, so we certainly should not see that mess again…right?
Unfortunately, the same brand of board that failed three times before failed again, was replaced, and failed a second time at the same transistor while using a “gentler” 1.40V. But this time ASRock wasn’t alone.
ECS’ P55H-A blew out what appears to be capacitors and didn’t even shut off afterward. We shut it down manually and the system still runs on two power phases, though we wouldn’t want to stress it. ECS’ failure occurred at the 0.29 V core voltage offset, which produced core voltage levels from 1.39 to 1.41 V depending on load.
The third try is supposed to be charmed, but it wasn’t for MSI. The P55-CD53 blew out one power phase at a mere 1.36V under full CPU load. However, the circumstances surrounding that failure are a little more interesting, since a 0.371V offset voltage was required to get the system to boot at 1.44V, where the load of eight Prime95 threads pulled its voltage down to 1.36V. It stayed at 1.36V for several minutes before the failure occurred.
Whom do we go to for answers here? Encouraged by a far-lower TDP rating on the new generation of Core i7 processors (95W, down from 130W), these companies probably designed LGA 1156-based models with a similar percentage of “overcapacity” as they’d used on LGA 1366 boards. But the fault doesn’t sit solely with each motherboard’s initial design team, as testing should have revealed the problem before the boards reached mass production or distribution.
Maybe a better question would be “what’s missing”? It seems that many of our previous-generation motherboards would shut off when overloaded, before anything was damaged. That’s called over-current protection, and it’s a feature apparently now reserved for high-end boards.
But did we push too hard? Remember that these settings were achieved using air cooling while keeping the CPU below its thermal-throttle threshold. Air-cooled overclocks certainly aren’t considered “extreme.” And for those wondering, the side of the VRM that burned in every case was the same side upon which the CPU fan was blowing strongly. This once again demonstrates that a voltage regulator should never be designed so that it can output more than its peak capacity.
Because three of the failures occurred at or below 1.40V, we have no knowledge of what to recommend as a “maximum safe” voltage for overclocking these three motherboard models. In the future, we’ll allow motherboard manufacturers to specify any lower “maximum” settings that may be required to preserve their products and will then duly arm readers with that information.
Update: ASRock confirmed that its P55 Pro was designed to handle a smaller power load, based on Intel's lower TDP for LGA 1156 processors, and has sent a BIOS its representatives claim will enable built-in over-current protection. To make sure that all three companies have an equal opportunity to address this issue, we have given them all two weeks to come up with a solution and will publish an article that addresses the new overclocking limitations.