AMD Unleashed: Four CPUs, Two GPUs, All Overclocked


In order to truly squeeze performance from each of these processors, we pushed them hard and were willing to use up to 1.55 V for the Phenom IIs if the Xigmatek cooler could keep load temps from going much above 50 degrees Celsius. Apart from pushing the northbridge frequency, the limits were already widely known for three of the processors.

When it was time to overclock the new Phenom II X3 720 BE, I eagerly started to raise the multiplier, hoping it would even surpass the high marks the Phenom II X4 940 BE set. But when stability testing at 3.4 GHz quickly resulted in a blue screen (suggesting that we had already exceeded the limits at stock voltage) hopes diminished into a hunch that this particular chip was not going to be as stellar an overclocker. 

Along the way, the Phenom II X3 720 BE required a voltage boost one-half multiplier (100 MHz CPU core speed) lower than the Phenom II X4 940 BE. For instance, reaching 3.6 GHz required 1.45 V with this chip versus the 1.40 V required for the previously-tested Phenom II. Final hopes of a higher-than-average overclock were crushed when the chip simply lost stability at 1.525 V and above. 

With a sweet spot of about 1.5 V, the chip initially passed 30 minutes of Prime95 stability testing at 3.7 GHz (18.5 * 200). In the end, we learned that it was not completely stable above 3.67 GHz (18 * 204). From there, the northbridge multiplier was raised to 12 resulting in a northbridge frequency of 2,449 MHz, and the HyperTransport (HT) was set back to its stock multiplier. The DDR2-1066 modules from Corsair did not require additional voltage for this small memory overclock.

In the CPU-Z memory screenshots, you may have noticed we were using unganged mode with each processor. Ganged mode is often the default BIOS setting and is considered to be better for single-threaded applications, while unganged can provide better performance in multi-threaded applications. Some initial testing was performed in both modes, with results indicating that neither really provided a large overall advantage. Unganged mode offered a slight edge in applications and games, while ganged mode took a few wins and provided much higher memory bandwidth scores in the Synthetic Sandra VII benchmark. So, while there may be specific applications for which one mode provides a real advantage over the other, overall, at least in our tests, effects are minimal.

As in our SBM series, the graphics cards were overclocked. We also sought to use the same GPU and GDDR5 frequencies that we used in the last round of $1,250 and $625 SBM PCs. The closest available GPU setting for the Radeon HD 4870 X2 was 782 MHz, which was 2 MHz higher than what Catalyst Control Center (CCC) allowed for the Sapphire Radeon HD 4870 X2. The memory was run at 950 MHz GDDR5, which was identical to the overclocked $1,250 PC’s memory speed.

However, this Radeon HD 4870 had a GPU core limit of 790 MHz in CCC, which was 15 MHz less than the Sapphire Radeon HD 4870’s core limit in the $625 PC SBM. Attempts were made to attain a clock speed of 805 MHz with Riva Tuner, and while the GPU didn’t artifact, the overclocking didn’t seem to stick, and performance occasionally throttled back to stock speeds during testing. In the end, we had to go with CCC, attempting to make up for the lower GPU frequency by running the GDDR5 at 950 MHz versus the 930 MHz with which our Sapphire card topped out.