We performed all our tests on a testbed built with the following components:
- Gigabyte GA-Z87X-D3H rev. 1.0 mainboard (LGA1150, Intel Z87 Express, BIOS version F5);
- Intel Core i5-4670K CPU (3.6-3.8 GHz, 4 cores, Haswell, 22nm, 84 W, LGA 1150);
- 2 x 8 GB DDR3 SDRAM G.Skill TridentX F3-2133C9Q-32GTX (2133 MHz, 9-11-11-31-2N timings, 1.6 V voltage);
- Gigabyte GV-T797OC-3GD (AMD Radeon HD 7970, Tahiti, 28 nm, 1000/5500 MHz, 384-bit GDDR5 3072 MB);
- Crucial m4 SSD (CT256M4SSD2, 256 GB, SATA 6 Gbps);
- Scythe Mugen 3 Revision B (SCMG-3100) CPU cooler;
- ARCTIC MX-2 thermal interface;
- Enhance EPS-1280GA 800 W PSU;
- Open testbed built using Antec Skeleton system case.
We used Microsoft Windows 8 Enterprise 64 bit (Microsoft Windows, Version 6.2, Build 9200) operating system, Intel Chipset Device Software driver package version 126.96.36.1997, AMD Catalyst 13.4 graphics card driver.
Our testbed has changed since our last review. First of all, we switched to using two DDR3 modules from the G.SKILL TridentX F3-2133C9Q-32GTX kit as system memory (the kit includes four 8GB DDR3 modules). By default, the memory is clocked at 1333 MHz with timings of 9-9-9-24-1T at a voltage of 1.5 volts. With the XMP profile, the frequency goes up to 2133 MHz at a voltage of 1.6 volts with timings of 9-11-11-31-2T.
Like most overclocker-targeted memory products, the G.Skill TridentX series is equipped with tall heatsinks. However, these heatsinks are composite. You can remove their top part by unfastening a couple of screws and make the modules smaller in height. This ensures their compatibility with more CPU coolers without compromising their operating characteristics.
It must be noted that the G.Skill TridentX series includes faster products capable of working at 2400 MHz. Of course, we checked this out but we couldn’t make our modules work at such a frequency even at higher voltage and relaxed timings. That’s why we’re going to use this memory at its default settings and with the XMP profile to test mainboards in default and overclocked modes, respectively.
Next, we replaced the Intel Core i7-4770K processor, the senior model in the Haswell series, with an Intel Core i5-4670K. We hadn’t been satisfied with the former in our review of the ASUS Z87-K as it could only be overclocked to 4.4 GHz. As opposed to its senior cousin, the Intel Core i5-4670K does not support Hyper-Threading, so it can execute four rather than eight instruction threads simultaneously. Its integrated GPU works at 1200 rather than 1250 MHz and the CPU itself is clocked at 3.4 rather than 3.5 GHz. We don’t often see it working at that clock rate, however, because the Intel Turbo Boost technology increases the frequency to 3.6, 3.7 and 3.8 GHz when four, three or fewer CPU cores are in use, respectively. The clock rate of the CPU-integrated North Bridge also changes at that, reaching 3.8 GHz. So, the two CPUs are similar in their clock rates, but the lack of Hyper-Threading might result in higher overclocking potential.
Our hopes didn’t come true, though. The Intel Core i5-4670K couldn’t reach 4.5 GHz, either. At 4.4 GHz, it only had lower temperature and power consumption compared to its senior cousin. We had to quickly return that CPU sample but instead we eventually got as many as three. The first of them turned out to be the best one in terms of overclocking potential, but we want to tell you about its less successful cousins first.
The three new samples of the Intel Core i5-4670K processor belonged to the same batch, which was different from the batch of the very first i5-4670K we had tested earlier. The second sample turned out to be the worst in terms of overclocking although its part number was similar to the first one. It must have had the highest default voltage. The ASUS Z87-K mainboard we used to test them didn’t show us the level of CPU voltage but we could estimate it indirectly thanks to the high precision of the CPU-integrated voltage regulator. Every Haswell processor we tested had a voltage of 0.708 to 0.718 in idle mode, but the voltage of the second Core i5-4670K sample was higher at 0.752 to 0.758 volts. So, it could not overclock even to 4.4 GHz at 1.2 volts. It might have reached higher clock rates at higher voltages, but we didn’t check this out since that sample was obviously no good for overclocking.
In the same way we gave up the third sample of the Intel Core i5-4670K. It was stable at 4.4 GHz but failed our tests at 4.5 GHz even at a voltage of 1.2 volts. The first sample, on the contrary, could work at 4.5 GHz with a rather low voltage of 1.150 volts. So, we hoped it could be overclocked to 4.6 GHz even. It did pass a lot of tests at that frequency, but we had to limit ourselves to 4.5 GHz to ensure full stability. The memory frequency was increased, too.
So it is impossible to pick up an overclocker-friendly Haswell by its batch number. One batch may include CPUs with high and low overclocking potential even if their part numbers are very close to each other. As for the Intel Core i5-4670K we’ve kept for our tests, we were quite lucky to make it work at 4.5 GHz at a voltage of only 1.150 volts, but we could only verify its ability to work at such settings with the Gigabyte GA-Z87X-D3H.
When we tested that CPU on our ASUS Z87-K, its clock rate would drop occasionally at default settings and, later, in overclocked mode. For example, running the Prime95 utility, the CPU would drop its frequency from 4.6 GHz to 4.4 GHz. When the frequency was set at 4.5 GHz, the actual frequency would drop to 4.3 GHz and then go up to 4.5 GHz again. It turns out that the ASUS Z87-K cannot really tell us that an overclocked CPU is stable or not. Moreover, during our performance tests we found out that the ASUS Z87-K had problems even when the CPU frequency multipliers had been manually set at their standard values. So now we wish we had started our tests of the new Intel platform with the Gigabyte GA-Z87X-D3H instead of the ASUS Z87-K.