Hands on: The need for speed

I then increased the external bus to 400MHz, which overclocked the processor to 3.6GHz and drove the memory at its maximum 800MHz specification. It booted fine and performed the encode in 146 seconds – interestingly, a little slower than the earlier 11x/333MHz configuration.

Increasing the external bus further overclocked the memory, at which point the system wasn’t totally happy, so I decided to stick with 400MHz and try it with a higher multiplier. With the bus at 400MHz and the multiplier increased to 10x, the processor was overclocked to 4GHz and booted Windows without issue. The encode was faster at 135 seconds, but only nine seconds quicker than the 9x/400MHz setting.

Sadly, that’s as far as I could overclock my system with conventional air cooling. Boosting the multiplier to 11x for a 4.4GHz overclock failed to post, and dropping back to 10.5x and 4.2GHz may have posted, but failed to boot Windows.

Power consumption
While overclocking my system, I kept a close eye on power consumption using a Maplin mains power meter. This revealed some interesting side effects that could change your long-term view on certain overclocking configurations. For consistency, I took a measurement at the same halfway mark on each video encode, when the load was at 85 per cent.

With the standard 9x/333MHz configuration, the system consumed 145W. This figure remained the same with the multiplier increased to 10x, but jumped slightly to 150W at 11x.

With the multiplier set back to 9x and the bus increased to 366MHz, the power consumption jumped to 155W. That’s 10W more than the 10x/333MHz configuration, despite both having the same final clock speed and exactly the same encoding time.

With the multiplier at 9x/400MHz, the consumption leapt to 172W – that’s 22W more than the 11x/333MHz configuration, despite both having similar final clock speeds, and the earlier one actually delivering a slightly quicker encode.

Finally, with the 10x/400MHz configuration, the consumption increased to 180W. That may be proportionately acceptable compared with the 145W starting point, given the 33 per cent boost in clock speed, but the surprising results were the differences in consumption when overclocking with the multiplier co mpared to the bus.

Multiplier magic
I’d expected the bus increases to yield the best results, given the memory was being driven quicker alongside an overclocked processor, but in this particular test increasing the multiplier alone delivered slightly better speeds. Of course, this is just one benchmark, and another that relies on memory bandwidth may tell a different story.

The really interesting result was the difference in power consumption. At 11x/333MHz for a 3.63GHz overclock, the encode took 141 seconds and consumed 150W at the 50 per cent mark. At 9x/400MHz for 3.6GHz overclock, the encode took five seconds longer, while consuming 22 more watts at the same point.

Clearly, overclocking the QX9650 by multiplier alone delivered the preferred result and, while that’s little consolation for those with locked multipliers, it does prove it’s worth performing additional tests to find the right balance for a long-term configuration.