One thing we quickly see from the temperature and power draw measurements is an exponential relationship. As you move left of the stock frequency, the power and temperatures don't drop by much, but as you go higher and higher from the stock frequency, we see an exponential, quickly-increasing trend. This is why overclocking can have such abrupt "caps," because the power draw and thermal output begin to increase very quickly. On the other hand, we see the practical limits of underclocking and undervolting, as you quickly begin to hit a point of diminishing returns, that is, as you push the undervolt or underclock, you are gaining less and less efficiency from the chip. How close the stock configuration is to that point says quite a bit about the efficiency of the stock frequency and voltage settings.
As we can see there is a fair margin of efficiency improvement that could be had from the stock clocks. This margin is there, of course, for a reason, which is universal out-of-the-box stability. The good news here is that he 1.00V undervolt was quite easy, and has proven to be completely stable. Voltages under 1.00V tended to be a bit pickier, but certainly attainable for even further efficiency gains. As we can see though, there isn't much margin from the stock undervolt to reduce temperatures or power draw any further, which is a great characteristic in the efficiency level at stock clocks. The 11.5W power draw reduction under load from the undervolt is a large percentage load drop. If a folder is running their system 24/7, that's 11.74KW-h of electricity. It's a fair amount, and the reduction in heat output is also pronounced, with a 4.25°C reduction under load, which is enough to affect cooling needs.
Another rather intuitive result is that the performance is proportional to the clocks, and independent of the voltage settings. This is because the clock frequency is the "speed" of the processor, which isn't changed by an undervolt, which removes any real downside (outside of stability) for a good undervolting chip.
Overall, we see that the i5-2300 is a breeze to overclock, despite not being one of the "k" overclocking models. We also see that the chip can be very efficient with a relatively quick and easy undervolt. This is a fantastic quality for silence/passive enthusiasts as well as distributed computing members who have their PCs running at 100% load 24/7. The undervolted processor at stock settings is very close to the point on the power and thermal curve where you begin to lose much of the benefit of underclocking or undervolting, as your gains are very minimal to the loss in performance or potential instability to the undervolt. The i5, overall, performed great across the board in a variety of real-world and gaming tests, and we confirmed that the undervolt doesn't harm performance in any way as long as the clock speed is maintained.
The super-efficient Sandy Bridge 32nm architecture is the root of the efficiency gain potential through underclocking, as well as its ridiculous overclocking potential. Modest overclocking feels too easy, and "tossing" an added 25% of performance or attaining a substantial underclock was far easier than we anticipated. The undervolted efficiency at 2.8GHz further validates the performance, value, and overall awesome nature that the Sandy Bridge family brings to the computing masses, just from the other end of the spectrum.
- << Prev