Overclocking, Cooling, and Power
The last bit of performance testing basically covers everything that doesn’t fall into CPU or onboard specific. That means idle and load power usage, temperature testing, and also overclocking. For power usage I ran the 4790k through wPrime to put it under load. This was with the GTX 780 installed so the difference between this number and the highest you can expect it to pull with the CPU under load plus onboard being used should be similar. Anyhow as expected the numbers were just under 10 watts higher under load and 2 watts at idle. With the increase in clock speed as well as the higher TDP I expected the 4790k to pull a little more power. Even so compared to the other CPUs tested it is still very power efficient.
Temperature testing was just to see how well the new thermal interface material works. As you can see below the load temperature dropped two degrees while running prime95. You can expect numbers to be higher when using the stock cooler. This testing was with a 120mm Noctua heatsink, but it does show a nice improvement, even before you consider that the 4790k is running at a much higher clock speed!
So with the better thermals and the added capacitors Intel seemed really excited for all of the press to put the i7—4790k to the test in overclocking. At Computex Intel even put together an overclocking competition pushing the limits of Devil’s Canyon both on air and with LN2. They managed to reach 5.5GHz on air and 6.3GHz with LN2, both are very impressive numbers. For our testing, I specifically wanted to see how it would compare to a Haswell CPU. I was worried that the increase in clock speed would just mean less room for overclocking. To give both an equal test I used Asus’s built in overclocking software to overclock each CPU exactly the same to prevent any inequality.
The end result was 4.5GHz with the i7-4770k and 4.7 on the i7-4790k. There is a little less headroom percentage wise but the actual overclock capabilities are obviously higher. Not to be outdone by a piece of software, I spent a little time toying with it myself and I was able to safely get 4.84 with a bclk of 103 and a multiplier of 47. Given the performance results from Computex I have a feeling that there is still room left in our sample, but I am more than happy with the results I did end up with.