Intel Pentium 4 3.2E

Approximately two years ago Intel launched their long awaited Northwood core. The Northwood allowed Intel to regain their performance crown from AMD in the high-end sector. The Northwood doubled the L2 cache of the existing Willamette core as well as shrunk the die from .18u to .13u. As time rolled by, the Northwood core underwent enhancements such as a faster system bus and platform upgrades, which have allowed us to reach speeds in excess of 3 GHz. After being a favorite among the gaming and overclocking community for over two years, the Northwood is nearing the end of its life; in its place will be the long awaited Prescott based Pentium 4 CPU that has been delayed since Q4 2003.

Finally it’s the moment everyone has been waiting for. Since the launch of the Athlon 64, Intel users have been touting that Prescott will beat out AMD’s 64-bit technology. Now that Prescott has arrived, can he dance as well as we've been led to believe?

Prescott

For today, Intel is launching a total of six microprocessors for the trusty ole Socket 478. Prescott will be available from 2.8 to 3.4 GHz with Hyper Threading and an 800 MHz front side bus. There will also be a lower-end 2.8A version without HyperThreading for existing 533 MHz FSB motherboards. On top of that, Intel will be releasing their current Extreme Edition Pentium 4 at 3.4 GHz and also the last Northwood core at 3.4 GHz. To differentiate the processors, all Prescott processors, with the exception of the 2.8A, will carry the E suffix; for example, 3.2E, 3.0E, etc.

In today’s article we will be taking a look at the Intel Pentium 4 3.2E.

Differences...

Die Shrink

The existing Northwood-based Pentium 4 was fabricated using a 130 nanometer process with a 146mm² die, 55 million transistors, that required around 1.5-1.55 volts. Prescott is based on a 90 nanometer process that shrinks the die size to 112mm², making the Prescott die 22% smaller. While the die size has shrunk by 22%, the transistor count has increased 2.5 times, coming in at 125 million. Since a smaller fabrication process allows engineers to squeeze more transistors into a smaller area, engineers were allowed to add more enhancements. With the 90 nanometer die shrink, the Prescott operates with a mere 1.3 volts and dissipates 103W of heat.

For the most part, a die shrink is beneficial as it allows more transistors to be crammed in a small area while lowering voltage requirements.

Extended Pipeline

To allow the Pentium 4 to achieve higher clock rates, Intel had to increase the length of the pipeline. With the Prescott, the pipeline has been extended from 20 stages to aproximately 31 stages. This would normally decrease the IPC (Instructions Per Cycle) of the processor, but Intel has other tricks up their sleeve.

Better Branch Prediction

To make up for the extended pipeline, Intel has improved the branch prediction unit of the Prescott. Taking a page from the Pentium M design team, an indirect branch predictor was added to the core for better branch prediction. Since the pipeline needs to be flushed every time a bad prediction is made, less mistakes will lead to fewer flushes, which maintains performance.

More Cache

Where the Northwood had 8KB of L1 cache that was 4-way associative, the Prescott doubles that number to 16KB, making it 8-ways associative. In addition to the increase in L1, the L2 cache has also been doubled from 512KB to 1MB. With more L2 cache, the system doesn’t have to fetch data from memory as often.