We didn’t expect any surprises during CPU and memory overclocking. If you remember, we managed to hit slightly higher CPU clock frequencies on a Gigabyte board that is why it would be logical to assume that Asus solution will fall a little behind in all performance tests in overclocked mode. At first it was exactly like that. We see a logical although not serious advantage of the Gigabyte mainboard in the first three tests that use the computational potential of the CPU to the utmost extent.
However, the data compression test let’s Asus solution get ahead of the competitor, and the gap between them is not that small at all. In the next SuperPi test Gigabyte board again falls behind Asus, although not that dramatically.
3DMark Vantage benchmarks show approximate parity, either of the two boards occasionally takes the lead here. However, in games, Gigabyte solution again falls unexpectedly behind.
There are a lot of mysteries in life, but this one was solved very quickly: the difference between the mainboard resulted from the QPI bus frequency. Remember what Clarkdale processor architecture is: it consists of two dies – one contains the CPU cores and another one - graphics and memory controllers and other components. And these dies are connected via the QPI bus. In nominal mode when the base clock is 133.3 MHz QPI bus frequency is formed with 44x multiplier and equals 5865 MHz. the BIOS of both mainboards sets this frequency automatically, but during overclocking Gigabyte mainboard tries to keep this setting as close to the nominal as possible, while the multiplier on Asus board remains at its maximum. As a result, during overclocking to 200 MHz base clock on Gigabyte mainboard, the QPI bus frequency turned out even a little below the nominal and equaled 5600 MHz with 28x multiplier. However, on Asus board at 195 MHz base clock QPI bus frequency and therefore the data transfer rate between the processor cores and the memory increased to 8580 MHz. only small synthetic tasks, where all data may fit into the cache do not depend on the speed of operations with the memory. Therefore, despite the lower processor and memory frequencies Asus solution yields a little to Gigabyte in purely computational tasks, but in applications that use memory more intensely, it outperforms the competitor quite substantially.
I have to point out that lowering the QPI bus frequency didn’t allow us to improve the overclocking on Asus board in any way. Later on, when we get to reviewing Gigabyte solutions based on the new chipsets, we will definitely check if they can overclock processors just as good at the increased QPI frequency. And at this point we would only like to repeat that it’s better not to leave any of the significant BIOS settings at Auto. In reality, the seemingly neutral “Auto” may in fact indicate either On or Off as well as the whole bunch of intermediate values between them. By the way, another proof of this point awaits us in the next chapter of our today’s review.
However, the performance difference between the two mainboards is noticeable, but that’s about it. This is the way it should be, since we were trying to ensure that both mainboards were running in similar testing conditions. In order to estimate the performance gain from CPU and memory overclocking more illustratively, let’s check out the following table, where we compare the results of Asus P7H57D-V EVO in nominal and overclocked modes side by side:
I am sure you will agree that despite all objective drawbacks of the new Clarkdale processors, they have one obvious advantage: excellent overclocking potential and impressive performance gain resulting from it. It makes sense to use Clarkdale processors in nominal mode only if you have very good reasons for it. For example, you don’t really care about the system performance, but prefer to have low noise and low power consumption. If this is the case, then maybe you should reconsider your CPU choice. An Atom based system will be much quieter and more energy-efficient, but without overclocking, you will only use part of the Clarkdale CPU potential.