STABILITY POWER TESTING
One of the keys to overclocking any processor is the guaranteed stability of your overclock. There are many programs that are designed specifically to stress the processor and overall system in order to ensure that the user has reached a point that can be considered stable.
Stability can be a highly subjective definition however. There is no one test that can be ran for X amount of hours that will guarantee stability of a CPU overclock in every scenario. A common misconception from many users is that they have to run some of these intensive stability testing utilities for 12 hours under the most extreme loads known to man.
Monitoring of the actual power draw during normal usage would yield some surprising results for those that have never checked their actual system usage. Loads over 60 Watts at the CPU are very rare, even under very demanding scenarios. This is where the stability testing angle comes in, when testing for stability we will see some extremely high wattage loads that you would never notice in real life usage, even if the users system is heavily overclocked.
We are using LinX and Prime 95 for stability testing, with 4 hour runs of each. Both processors are run at the stock setting, and then overclocked to 5.0. Note that we are not testing overall system draw, just the CPU socket.
The power differences between our 3930K and the 3820 are tremendous when monitored during our stability testing utilities. The first sets of tests were conducted using stock settings for both processors. Being a six-core processor the 3930K does suck quite a bit more power. We expect the power draw to be roughly 35 percent higher for the 3930K, as that would be a linear increase representative of the difference in core count. The loadings for the 3930K under stability testing when overclocked are very high, among the highest power draw we have ever witnessed.
When taken into consideration with the power draw that is measured, one can note that the 3820 with a very hefty overclock of 5.0 draws as much power as the 3930K at its stock settings. From a single core perspective, the 3820 will provide the same, or slightly faster, speeds than the 3930K.
IDLE POWER TESTING
A key consideration when using a computer is the idle power draw. The vast amount of time that the average user is at the desktop or browsing the internet, the CPU is either at idle, or very close to idle. Above with our stability testing we highlighted the performance under 100 percent load in unrealistic scenarios, and here we will look at the performance at closer-to-reality idle speeds.
SpeedStep is a great feature that is designed into Intel CPUs to adjust the speed of the processor down to very low levels while idling. This can create drastic power savings over the course of time, and also has the great side effect of a quieter, cooler system for the end user.
One of the greatest things that we have observed during our testing of the Sandy Bridge CPUs is the fact that SpeedStep works very well when the processors are overclocked. With the X58 and 775 series of CPUs you could either have SpeedStep or Overclocking, but not both. The SpeedStep functionality always seemed to introduce too many stability issues in ‘bleeding-edge’ overclocks.
The overall power consumption of any consumers’ computer, with the exception of it being used as a 100 percent high-load server, folder, or cruncher, will be affected more accurately by the idle power draw. Being wonderfully well-integrated into the SNB-E series, we decided to monitor idle power draw with SpeedStep enabled and disabled. We are also comparing the power draw with both stock and overclocked configurations.
The stock power draw of both the 3930K and the 3820 are very close when SpeedStep is enabled, with only a difference of 16 percent. With SpeedStep disabled the gap widens a bit to 20 percent.
When the idle power draw is compared with a 5.0 overclock, the 3820 draws an extremely impressive 25 watts! This is very good, with a whopping 28 percent increase in clock speed only costing a miserly 16 percent more idle power.