Intel Presler: Lowering the Production Costs

The first generation of Intel’s dual-core CPUs based on 90nm Smithfield core had a lot of drawbacks. This is actually not surprising at all, because the decision about the launch was made without the proper preparation. As a result, Smithfield processors made from Prescott cores turned out not very fast and economical, and at the same time appeared pretty pricy to manufacture. Of course, the die size of the Smithfield based CPU is 206sq.mm, which makes it very hard to achieve high production yields even with the polished-off manufacturing technology.

When working on the Presler design, Intel developers did their best to eliminate at least the major drawbacks of the dual-core Pentium D processors. Especially, since there was every reason for that: the upcoming launch of the finer 65nm production technology. However, you should keep in mind that it is impossible to significantly improve Presler core, because Intel doesn’t have any other processor architecture besides NetBurst at hand right now. As for NetBurst architecture, it cannot be improved as well, as we have already mentioned above. So, you should regard Presler as Intel’s best attempt to improve the Smithfield core without investing a lot of engineering resources into it.

The first simplest means of improvement in this case is certainly the new production technology. The introduction of the more advanced manufacturing process allows reducing the processor die size on the one hand and increasing its clock frequencies (or reducing the heat dissipation and power consumption) on the other. Since the typical heat dissipation of the top Intel Pentium D processors based on 90nm core has already been increased to 130W, it doesn’t make much sense to reduce it back again by shifting to the new production process. The entire infrastructure has already been adjusted to support this pretty high heat dissipation value. So, the new 65nm dual-core Presler processors will boast higher clock frequencies than their predecessors instead. While the maximum clock rate of the Smithfield based CPUs equaled 3.2GHz, Presler processors will be able to reach 3.46GHz bar.

Thanks to the new production technology Intel managed to increase not only the clock speeds, but also the number of transistors in the CPU. As a result, the L2 cache memory has grown bigger. While Smithfield CPUs featured two L2 caches, 1MB each, i.e. were composed of two regular Prescott cores, the new Presler processors acquired a twice as big L2 cache. So, Presler can actually boast two L2 caches, 2MB each. As a result, it actually means that Presler is in fact made of two Cedar Mill cores, which can also be regarded as 65nm analogs of the Prescott-2M cores used in Pentium 4 600 processor family.

Higher clock frequency and larger cache memory are actually the only two things Intel did to increase the performance of its new-generation dual-core processors manufactured with 65nm process. There are no other in-depth architectural improvements in Presler whatsoever. Therefore, the performance of Intel Smithfield and Presler processors working at the same clock speed may only be different if the application is sensitive to the size of L2 cache memory.

To prove this statement by some practical results, we decided to run a synthetic SiSoft Sandra 2005 benchmark, which is known to use pretty simple algorithms, which do not get affected by the size of the available L2 cache memory. We had a Presler CPU with the unlocked multiplier and enabled Hyper-Threading technology (Intel Pentium Extreme Edition 955), which we decided to compare against Smithfield (Intel Pentium Extreme Edition 840) in the same testing conditions. So, we set the frequency of our Presler CPU to 3.2GHz (16 x 200MHz).