Overclocking
Completely opposite of the $2,500 machine’s ease of assembly was the difficulty of its overclock, calling into question our choice of a Core i7-860 processor and a mid-priced motherboard.
The problem was heat. We did a great amount of testing on this particular Core i7-860 processor and found that it could actually exceed the overclocking capability of the Core i7-870 used in our P55-UD4P motherboard review. However, the i7-860 also converted 10% more power into heat (330W compared to 300W) at 1.45V core and full load. We thus needed more cooling.
Unfortunately, we also found that the Xigmatek HDT-S1284 cooler produced similar temperatures at 270W as our reference cooler produced at 300W. Further inspection of the HDT-S1284 revealed several potential design problems, including a low-speed fan that couldn’t be increased regardless of its PWM-control capabilities, a sink that leaked most of that fan’s air from the side, leaving the center of its fins almost unused, and an add-on installation kit that provided little contact pressure. We tried resetting the fan several times but got our best results (a drop of nearly two degrees Celsius) by simply pressing down against the sink’s top.
Given a 10% higher thermal load and 10% less cooling capability, we weren’t going to get anywhere close to the processor’s limit at our chosen voltage. Our new limit would be 1.35V, with 4.0 GHz and a full CPU load of eight Prime95 threads pushing the CPU to 90 degrees Celsius.
We then encountered a second heat issue: at full GPU load, the graphics cards would pump too much heat into the case. This would have made it impossible to load both the CPU and GPU simultaneously, a condition we certainly lament.
Using a BIOS setting of 1.30625V with load-line calibration enabled, our CPU would climb to 1.312V but only clock to 3.65 GHz stably, with both the CPU and GPU cores at full load.
Knowing that the system could potentially support even higher clock speeds when only one or two threads were enabled, we also tried getting to this point using Intel Turbo Boost. A base clock of 162 MHz got us to 3.56 GHz with for cores loaded, 4.05 GHz with two cores active, and 4.2 GHz with a single core active, while the slightly lower four-core results yielded to better average performance thanks to how many of our benchmarks are single- or dual-threaded. An added benefit was lower idle power, since Intel’s power-savings features must be activated in order to reach the highest Turbo Boost ratios. Yet this turned out not to be the perfect solution we’d hoped for, as core voltage occasionally didn’t increase fast enough at program launch to keep the system stable. Our demands for perfect system stability forced us to revert to old-fashioned low-efficiency overclocking methods.
Our RAM was capable of reaching the same DDR3-1600 CAS 8-7-7-18 timings as the previous set, but only when a fan was over it. That could be partly due to P55-UD4P BIOS increments of 0.02V, which forced us to set 1.66V rather than 1.65V, but it’s also true that this month’s $2,500 machine lacks the accessory fan used in September. To keep things cool, we decreased DRAM voltage to 1.64V, a drop that along with the slight increase in clock speed, forced us to use looser 8-8-8-18 timings.
Our graphics card BIOS limits GPU overclocking to 900 MHz. DRAM restrictions were more adequate, allowing us to reach the memory’s stable 1,270 MHz limit. We know that these cards could go much faster using unlocked BIOS and another manufacturer’s Afterburner Utility, but we didn’t want to void the graphics warranty of this future giveaway system by flashing non-native card BIOS.
Tuners interested in copying these efforts (or using them as a baseline) can view BIOS screenshots by clicking on the above thumbnail images.
