Given the previous two pages of overclocks, performance results, and BIOS settings, it'd be easy to assume that our previous results were a fluke. Adding the aforementioned observations to ASRock’s advantageous slot layout could make the X58 SuperComputer a winner in many minds.
But the preceding tests don't tell the whole story, nor do they entirely address the cause of our previous component failures. We needed to get to the bottom of the matter, if only to provide builders some peace of mind about what they can expect
If only to back up the findings from the boards provided by ASRock to test and re-test, we also purchased a retail board from Newegg to replicate earlier testing from our Sub-$300 X58 Motherboard Comparison. Intel even sent a replacement C0-stepping processor with the same batch number as the part we lost. Testing on other boards showed it has the same overclocking capabilities and voltage requirements of the unit it replaced.
Two minor changes were required to replicate real-world test conditions however. First, the PCB revision we received from Newegg was 1.03 (versus 1.04, which we used in the published motherboard comparison). High turnover keeps Newegg’s products ahead of most competitors, so this is probably the most recent version retail buyers will find for at least a week or two (we confirmed that Newegg has 43 boards in stock as of this writing). We also used the most recent BIOS version, since this is one of the first items on the checklist of many real-world overclockers.
Unfortunately, using our "review" setup (with the retail C0-stepping Core i7), we encountered the same issue, where keying in too-high of an input voltage, with the aim of achieving an ample output voltage under load, would result in overloading the VRM. And “Vcore Loadline Shallow_SLOPE” setting (not available on the rev. 1.03 board) does have something to do with it, as does our use of a C0, rather than D0, Core i7 processor.
Answering the "Why?"
Most C0 processors require more voltage to reach their overclock ceiling compared to later D0 parts. The added voltage puts more force against the circuit resistance, generating more heat, and consuming more electrical current (amperage). Among other things, an electrical formula known as Ohm's Law proves that if resistance is held constant, amperage increases in direct proportion to voltage. Our 1.45V target voltage would put a significantly higher load on the motherboard’s voltage regulator compared to the D0 core’s 1.416V goal, assuming that there weren’t any significant changes in material or design between core revisions.
Booting PCB Revision 1.03 at a 1.45V Vcore setting and a moderate 3.80 GHz overclock resulted in instability, with a 1.38V output reading at idle dropping to around 1.32V under load. With no automatic compensation for core voltage droop under load, we attempted to compensate manually by choosing a setting that would be closer to 1.45V output at idle. A BIOS setting of 1.50 volts provided an actual output voltage of 1.44 volts at idle, but this dropped to 1.36 volts under full CPU load. This 1.36V output was enough to keep the C0-stepping Core i7 stable at 3.9 GHz, at least for a quick test. But we wanted more because we knew the processor could do more.
Curious to find a setting that would yield the preferred 1.44 to 1.45 volt Vcore under full load, we used ASRock OC Tuner to find it at a 1.60V manually-keyed-in voltage, then backed down immediately prior to additional testing. A lower specified voltage would help to prevent damage when voltage spiked upon relieving the load, or so we thought.
Next would be the attempt to reach 4.0 GHz, a frequency commonly reached with our C0-stepping processor core. With an output voltage of 1.40V at full load, we felt safe trying the 1.56V setting for the 4.0 GHz attempt. After less than ten minutes under Prime95, the e-tail-purchased revision 1.03 motherboard failed. Fortunately, our CPU survived this time.
Fortunately, there are several ways to prevent anything bad from happening to your setup, if you also have a 1.03 board and C0 processor. Read on, for all is not lost here.