Sign in with
Sign up | Sign in

CPU Overclocking

System Builder Marathon, March 2011: $2000 Performance PC
By

We borrowed a bridge from another motherboard for our tests, but buyers who choose to instead assemble the system with included components can use a closer slot for the second card, at the same time reducing airflow to the card above it and knocking both slots down to x8 mode. Finally, anyone who upgrades to three-way CrossFireX will be forced to use the closer slot for their middle card, and the CrossFire bridges included with the graphics cards are long enough to reach from the first to the second and from the second to third PCIe x16 slots.

While “Sandy Bridge” overclocking is primarily limited to the manipulation of its Turbo Boost multipliers (on unlocked K-series models), the process is actually fairly simple. We locked in a 100 MHz base clock and increased the maximum Turbo Ratio to 48x, set Load-Line Calibration to High, and used an Offset Voltage of +0.100 V to push our voltage under load to 1.40 V without disabling power-saving features.

Remember when we said we thought that G.Skill’s less-expensive, 1.5 V DDR3-1600 CAS 8-8-8-24 likely used the same components as its 1.6 V DDR3-1600 CAS 7-8-7-24? Increasing our 1.5 V memory voltage to 1.6 V allowed us the choice of 7-8-7 timings at DDR3-1600 or 8-9-8 timings at DDR3-1866, and our calculator told us the later setting would perform better. Hint: 1866/8 * 7 = 1633, and 1633>1600.

Here’s where we can pick between locked-in and floating multipliers, though it doesn’t list these options in UEFI. Instead we see “C1E, C3 Report, and C6 Report” options, and disabling all three will force the CPU to run at its maximum Turbo Boost ratio full-time. The EIST setting does not produce similar results.

Because power management does not reduce the performance of most applications but does save a significant amount of energy on-average, we left these settings enabled for our initial tests. After later finding that these settings do reduce our hard drive throughput tests, we wound up re-benchmarking the system with C-States disabled. Both data sets are represented in today’s charts so that readers can make an informed decision regarding the importance of this feature.

React To This Article