Additional Tweaking

With the Phenom II X3 710 humming along at a respectable 3,744 MHz, it was time to turn our efforts to squeezing even more performance out of this system.

We started by overclocking the northbridge, which increases memory controller performance and L3 cache speed. Setting the CPU-NB Voltage to 1.3V and the NB Voltage to 1.25V allowed us to increase the northbridge multiplier from 7x to 9x, resulting in a northbridge speed of 2,592 MHz.

Increasing voltages a little more still didn’t allow the system to boot into Windows at a 10x NB multiplier. Keep in mind, though, that at a 288 MHz reference clock, each step up in NB multiplier increases the northbridge speed by 288 MHz. The chipset heatsink was still fairly cool to the touch, but reaching a 2,880 MHz northbridge speed would require more CPU-NB voltage than we were willing to feed. This is another area where a Black Edition processor is more flexible. Using a combination of multiplier and reference clock overclocking could allow us to dial in a higher northbridge speed and very similar core speed. For instance, at a 270 MHz reference clock, the system was fully stable running the northbridge at 2,700 MHz, but without being able to increase the CPU multiplier, the CPU would only run a pinch over 3,500 MHz.

While it’s possible to see a small increase in performance by increasing the HT Link speed, 2.0 GHz generally provides plenty of bandwidth for a system like this. Here again, increasing the HT multiplier to 8x would add 288 MHz to the HT Link speed, resulting in 2,304 MHz, which is higher than we would normally attempt to utilize and would likely result in a loss of stability.

Instead of putting time into increasing the HT Link, we turned to overclocking the RAM. In this case, increasing the memory ratio to 1:3.33 would result in our Corsair DDR3 running at an unhealthy 1,920 MHz, so we instead focused on tightening memory timings. We found 7-7-7-20 to be fully stable in Memtest 86+, Prime95, and 3DMark Vantage testing. Unfortunately, while a 1T command rate passed four loops of Memtest 86+ without any errors, it resulted in a loss of stability in 3D testing. Details for the final “Tweaked OC” can be seen it the following CPU-Z screenshot.

Although we manually set memory timings for today’s overclocking, we found through additional testing that the Auto settings did not impact the overclocking efforts in any way. At a 1:2.66 memory ratio, setting the DRAM Timing mode in the BIOS to Auto resulted in 9-9-9-24 timings. Interestingly, we found that Auto timings with a 1:2 memory ratio resulted in 6-6-6-15 timings that proved stable with a 1T Command rate at this frequency.

For performance testing, we’ll separate these tweaking efforts, first seeing what performance is gained by increasing the northbridge speed alone and then taking a look at whether memory speed and timings have any additional impact on performance.