Sign in with
Sign up | Sign in

CPU And GPU Overclocking

System Builder Marathon, August 2012: Alternative $2000 Gaming PC

In order for us to pull off a perfect performance comparison between today’s GPU-heavy configuration and Wednesday’s CPU-heavy build, we’d first have to count on getting both systems to run at the same speed. That shouldn't have been a problem for the less complex Ivy Bridge-based Core i5, since our goal was a fairly ordinary combination of a 4.6 GHz processor clock rate and DDR3-2133.

Unfortunately, that perfect comparison still relied on one variable we lacked this time around: luck. After hours and hours of testing at ever-increasing CPU core voltage levels, we kept finding occasional bugs that appeared only after we thought we were running stably. Our new mark turned out to be 4.5 GHz.

We were similarly unable to get our memory to run at DDR3-2133. Starting with the kit's XMP profile, we continuously increased the memory multiplier and watched as both platforms loosened memory timings accordingly. Yet, the previous system’s ASRock board appeared to choose more appropriate tertiary timings than Gigabyte’s, and our manual experiments with tertiary timings lead to naught.

We were able to use the same 10-11-10-24 timings as the first $2000 system's, but at DDR3-2000 data rates instead of the previous build's DDR3-2133. At least we didn't have to mess with secondary and tertiary timings.

Starting at our target 1.25 V core, we increased voltage up to 1.32 V in an effort to reach 4.60 GHz before giving up, setting 4.50 GHz, and finding stability at the same 1.25 V we originally picked. Our DRAM required 1.65 V.

Though this specific CPU was probably responsible for most of our overclocking issues, we did find noticeable voltage instability on the G1.Sniper 3 motherboard. Increasing its Loadline Calibration level to Extreme kept it above 1.25 V, but we’d really rather use a board that maintains a stable voltage at our desired level.

A maximum GPU clock of 1215 MHz with GDDR5-6610 memory again lowers the bar compared to our previous single-GPU build, but we expect that from an SLI array. After all, we're messing with two cards simultaneously, and the worst overclocker limits the ceiling for the entire configuration. That upper limit is further lowered by more heat, electronic noise, and the other stresses involved in running two graphics cards concurrently.

React To This Article