Overclocking

We haven’t yet undertaken any overclocking experiments for dual-core CPUs. Nevertheless, we couldn’t leave out this exciting aspect, because there will definitely be enthusiasts among you who would wish to squeeze even more speed from their dual-core processors. Especially, the youngest processor model of the Pentium D family is simply calling for an overclocker: it just has to overclock at least up to the level of the top processor in this product line, which works at 3.2GHz. Moreover, keeping in mind that Smithfield is a combination of two Prescott cores of E0 stepping on a single semiconductor die, we can expect even more fruitful overclocking: the latest CPUs based on Prescott and Prescott-2M core proved easily overclockable beyond 4GHz bar. However, dual-core processors should have more specific and sophisticated overclocking procedures than their single-core fellows. The main problem will inevitably lie with their high heat dissipation. All in all, we couldn’t help carrying out a few practical experiments.

For our Pentium D 820 overclocking experiments we assembled a platform on ASUS i955X based P5WD2 Premium mainboard. The cooling was provided by a pretty efficient Zalman CNPS7700Cu cooler. During overclocking we didn’t increase the processor Vcore. The CPU was overclocked by raising the FSB frequency: the clock multiplier of the Pentium D 820 processor is locked just like by Intel’s single-core solutions. The PCI Express and PCI bus frequencies were fixed at the nominal values. The memory frequency was reduced a little bit so that the Corsair CM2X512A-5400UL memory modules didn’t turn into a bottleneck during overclocking.

The actual overclocking of Pentium D 820 processor started quite unexpectedly. We managed to raise the FSB frequency to 250MHz in no time and without any special effort. This brought the CPU clock rate up to 3.5GHz.

The system started and booted-up normally at this speed and allowed running all sorts of small utilities (including single-threaded ones) without any problems. However, when we took a closer look at the system functioning in these conditions, we saw that it could not provide fully-fledged functionality in this case. The Zalman CNPS7700Cu cooler couldn’t cope with its task when the CPU got loaded heavily.

The temperature kept growing and the CPU would shut down. Since Zalman CNPS7700Cu is a pretty powerful air cooler, the conclusion we can draw here is not very optimistic. You need an extremely efficient cooling solution, a water-cooling system, for instance, if you intend to try overclocking your dual-core processor. We still decided it would be interesting to find out what is the maximum frequency our processor can work at stably with the air cooler, because this type of cooling is currently used in most computer systems. We reduced the FSB frequency to 240MHz and continued out stability tests. The CPU was working at 3360MHz in this case.

The results obtained in this case turned out even more interesting. The thing is that the system remained stable, the CPU temperature stayed within the acceptable range, and all major tests would run smoothly. However, when we started a benchmark supporting multi-threading, some weird things began happening. The results appeared suspiciously low in this case. When we launched two independent copies of a single-threaded application (we used WinRAR archiving utility), we managed to reveal a very interesting effect: Pentium D cores can get into thermal throttling independently. In other words, the first core continued working at its normal speed, while the second core started missing clocks losing its speed tremendously. The strange thing is however, that the temperature still remained within acceptable range of 75-78oC.

It turned out that the explanation is fairly simple. The temperature reported by the CPU is the temperature of the first core only. Unfortunately, Pentium D processors do not allow monitoring the temperature of both cores, this feature is implemented only in dual-core server processors.

So, to our great disappointment we cannot find out what actually happened with the second core. In our case, only the second core got overheated and went to thermal throttling, while the first core kept working normally.

So, we had to drop the FSB frequency to 230MHz and continue our tests. The CPU in this case worked at 3220MHz. At this frequency the second core didn’t get overheated any more, which is actually not a surprise for us, because this frequency is close to the clock rate of the top Pentium D CPU models.

So, our experiments suggest that Pentium D can only be overclocked if you have efficient cooling solution at hand. Regular air coolers allow significantly modest overclocking of this processor up to 3.2-3.3GHz only without thermal throttling of any of the cores.