Investigations into Athlon X2 Overclocking

System Settings

One of the key factors for a successful overclock is choosing memory timings and speeds that will work. It is also necessary to tweak the HyperTransport speed and multiplier. We took screenshots in CPU-Z showing the CPU and Memory tabs, which will give some details on each configuration that we benchmarked. The images from CPU-Z are available in a single Zip file (1 MB) if you want to see them (they will not be shown elsewhere in this article).

Most of the changes that we made in the BIOS are pretty simple, and while DFI provided the option to tune many memory timings, we stuck with the options that are available on the majority of enthusiast motherboards. We modified multipliers, RAM ratios, voltages, and the standard memory timings: CL, tRCD, tRP, tRAS, and CMD. All other memory timings were left at Auto. If you have a motherboard that doesn't support the same settings that we used, you will need to use some trial and error to discover the optimal settings for your board.

Click to enlarge.

There are a ton of variables involved at each tested setting, and stability and settings are going to be different for each set of parts. We could have tried for more optimal settings, but the amount of time spent running benchmarks is already huge, and considering the last time, we'll leave additional performance tweaking as an exercise for the reader. Patience along with trial and error are required for any extended OC attempt.

Note how CPU voltages scaled rapidly as we neared the highest overclock levels. We did spend more time checking out voltages for this article, and unless a configuration could run through both 3DMark and both PCMark tests in sequence without crashing, we did not consider the voltage stable. We increased the voltage in 0.025V increments until the Futuremark tests all ran to completion.

While the Venice chip that we tested was fully stable at 2.7 GHz and slightly flaky at 2.8 GHz, the Manchester core that we used was slightly less capable. 2.6 GHz presented no problems at all during testing, and we ran Prime95, SuperPi, Folding@Home, and many games with the system running at 2.6 GHz without any crashes. 2.7 GHz was a different story, with periodic crashes in several games, particularly after an extended period of time. Prime95 and Folding@Home also experienced problems with the 2.7 GHz clock speed, though SuperPi still managed to calculate 32M decimals simultaneously on both cores. Other X2 chips may perform substantially better, or possibly worse, and better CPU cooling may provide a solution. (We've heard of some 2.8+ GHz overclocks - we could barely get into Windows at 2.8 GHz, and even sitting idle the PC would still crash.)

Again, this article presents results merely as a guideline, and depending on the components used, you will almost certainly need to change a number of settings in the BIOS. A crash in many ways is the best result possible - it means that you are definitely unstable. While passing the entire benchmark suite suggests that a system is stable, there is still no guarantee that a future application won't cause problems.

RAM timings and speed are the other items that we changed quite a bit. We kept the value RAM at a constant 2.5-3-3-7-1T 2.6V setting while adjusting the ratio to keep the speed at or below DDR400. The PC-4800 RAM only required minor adjustments to reach up to DDR-540, and we left the voltage at a constant 2.8V (2.75V was the BIOS reading - the RAM voltage setting was always around 0.05V higher than the measured value). Voltage for the VX was the maximum 3.2V that the motherboard could supply, and even then, it was insufficient at higher bandwidths to allow 2-2-2-7-1T timings. Something like a DFI LanParty should be able to achieve even better timings at overclocked settings with a 3.5 V setting, though OCZ VX and Mushkin Redline are becoming difficult to find, so perhaps it's a moot point. Finally, the Patriot 2GB sticks were good for up to DDR-480 speeds, give or take, with 2.5-3-3-8-1T timings. 2.7V appeared to provide the best results, with 2.8V perhaps helping slightly at the highest RAM speeds. The PDP RAM was able to remain slightly ahead of the value RAM in terms of timings and speed, while of course offering twice as much RAM.

One area where we experienced serious problems was in running with four DIMMs. First, as expected, a 2T command rate was required. Even then, using four OCZ EL Platinum DIMMs, we were only able to complete benchmarks at 2-3-2-7-2T timings (2.8V) at default CPU clock speed. All attempts to get the system to run stable in any overclocked state of 2200 MHz or more met with failure. Winstones could run successfully in almost every instance, including 2.5-4-4-8-2T timings at 2700 MHz, but Battlefield 2, Far Cry, and FEAR all crashed (hard-locked the PC) repeatedly. Test #5 in MemTest86 also failed with display corruption. Attempts to tune the RAM better met with limited success. By changing the EQ Drive Strength to "Weak", we were able to get further in testing, but nothing was truly stable. We would venture to say that the RAM is not to blame, as both pairs of DIMMs worked separately. The most likely culprits are the motherboard and BIOS, which are not as highly tuned as many enthusiast motherboards.

A last comment is that we didn't fully benchmark all of the settings listed in the charts. We tested 2000, 2200, 2400, 2600, and 2700 MHz. In order to provide a linear scale (so that the results at 2700 MHz aren't skewed), we interpolated the in-between scores. This is a problem with the graphing capability that we have within Excel. Basically, ignore results at 2100, 2300, and 2500 MHz - you could probably get scores within a few percent of what shows up on the graphs, but we didn't actually verify this.