Performance

Asus P7P55D Deluxe mainboard is the first LGA1156 solution that we have tested in detail in our lab. It doesn’t yet have any competitor we could compare it against that is why let’s see what performance gain we can get from overclocking Intel Core i7-860 processor. In nominal mode the board set all parameters on its own, we only enabled dynamic Turbo Boost implementation manually. At first let’s compare the performance numbers with those obtained during overclocking when the clock frequency multiplier also changes dynamically.

The numbers are quite impressive. If we disregard those cases when the performance is limited by the graphics card, the gain makes about 20-30%. And now let’s check the performance gain compared with the overclocking results obtained with static Turbo Boost implementation in place.

This time, the performance gain from overclocking reaches 40% in some cases. And finally, let’s compare the performance numbers obtained during overclocking using static and dynamic Turbo Boost implementation:

What we see is really puzzling. If graphics card becomes a limiting factor, the performance numbers are about the same, and in almost all other cases we see dynamics falling about 5-10% behind statics. And although on average the lag makes only about 4.5%, it doesn’t really work as a consolation, as we expected the dynamic mode to win! Formally there is nothing surprising about it. Our benchmark set is put together for the purposes of comparing mainboards, and not CPUs. Moreover, we have specifically selected mostly multi-threaded applications that can use the potential of multi-core CPUs. So, what equal conditions are we talking about if the CPU works at 3.55 GHz with dynamic implementation and at 3.9 GHz with static one? Of course, statics is faster. The only single-threaded application that we have added to our benchmarking suite in order to estimate the possible gain from dynamic Turbo Boost technology implementation is SuperPI. This is where we naturally see a logical advantage of the dynamic approach.

So, we started to look frantically for single-threaded applications that could convincingly demonstrate the advantages of dynamics over statics. To my great surprise, I didn’t find anything like that. Of course, we can run Cinebench or Fritz with only one thread and get the desired result, but it has nothing to do with the real state of things. I doubt that anyone will give up multi-threading and sacrifice higher performance only in order to raise the CPU frequency. We only care about maximum speed and it doesn’t matter how we achieve it: by increasing the frequency or the number of simultaneously executed computational threads. If the second approach is much faster, no one will resort to the first one. This is where a paradoxical at first glance conclusion comes to mind: static Turbo Boost implementation performs much higher during overclocking than dynamic one.

In fact there is nothing surprising here, dynamic Turbo Boost implementation shows all its advantages only when the CPU works in its nominal mode but not during overclocking. So what changes when we switch from statics to dynamics in nominal processor mode? Nothing, except the fact that in some cases we allow the CPU to increase its own clock speed. We have the same base clock of 133 MHz, which means that all connected busses, such as the memory bus, for instance, have the same frequencies. Of course, in this case dynamic implementation is preferable, as we can see from the results of our comparison. We see a convincing and logical advantage of the dynamic implementation when the CPU works in its nominal mode.

And when we switch from statics to dynamics during overclocking, everything changes. We had to lower the base clock and as a result all frequencies tied up to it lowered, too: by lowering the base frequency from 177 to 161 MHz we automatically reduced the memory clock from 2124 to 1932 MHz. Of course, more aggressive memory timings partially make up for this lowering, but nothing will be able to cover up the drop in the CPU frequency under heavy load. Yes, sometimes the CPU frequency will increase to 4.2 GHz, which is higher than 3.9 GHz with static Turbo Boost, but at the same time it will often be only 3.55 GHz instead of the same 3.9 GHz. Keeping in mind that there are barely any contemporary single-threaded calculations these days, because the CPU anyway has to pay some attention to OS and other applications’ requests, it appears that we only get maximum performance during overclocking with static Turbo Boost implementation. Of course, we could occasionally run Pi calculations just for maintaining our self-esteem with dynamic Turbo Boost used during overclocking, but it is hardly practical. We could also dig out some old single-threaded games, where we will also see a performance increase, but the performance of contemporary processors and graphics cards is more than enough for old games even without the Turbo Boost. All in all, dynamic Turbo Boost implementation turns out to be less useful during CPU overclocking than static one.