Prescott v1.1
The new Pentium 4s are based, like the previous versions, on the Prescott core, which was introduced almost a year ago. This controversial processor has a 90 nm fabrication process, and even if it has twice as many transistors as the previous version isn’t faster in practice.
In fact, most of the transistors were used to compensate for performance losses due to architectural modifications such as longer pipelines, or the increase in cache latency, which were supposed to allow a rise in frequency for the NetBurst architecture. The new Pentium 4’s power consumption and heat dissipation was higher than the previous version, and still didn’t provide any significant performances improvements.
Intel is now back with an evolution of the Prescott core. Three main innovations have been introduced compared to the Pentium 4 5xxJ, which was released in November (see this test) :
- The cache L2 has been increased from 1 MB to 2 MB.
- supports Enhanced Intel SpeedStep Technology (EIST)
- and also Enhanced Memory 64 Technology (EM64T)
The first innovation speaks for itself and means processor size has clearly been increased. The number of transistors has jumped from 125 to 169 million. It’s also important to measure the results of this new cache in practice. Here are the performances obtained.
Cache latency time is similar, but higher than the Northwood. The transfer rate is the bad surprise: Cache L2, whose transfer rate has already gone down with the transition to the Prescott core, has once again been reduced. In fact, it’s as fast for the first 256 KB, but is noticeably slower for the rest.
The EIST is an energy saving function similar to the one integrated to laptops. Depending on the level of use, the Pentium 4 6xx coefficient could be reduced by as much as x14, and electrical tension to1.2V. Since the J version the Pentium 4 already features a power management function to reach the same level with the Enhaced Halt State (C1E). With the C1E, it was either x14 or the maximum coefficient without intermediate levels. So at 0% of use the coefficient was x14 / 1.2v, and for a use of 5%, the coefficient and electrical tension were set to the maximum.
With EIST, depending on the level of use the coefficient and electrical tension varies. It’s interesting to note that for the moment this function isn’t natively supported by Windows, because of a lack of adequate drivers. A utility like
RM Clock can manage this function. You may notice on the bottom of the screen that the coefficient now isn’t the minimum or maximum depending on the level of use. These factors vary according to use unlike the screen above. Electrical input is actually reduced to 1.2v in stand by. With C1E, however, and we can’t explain it, the value varies between 1.237 and 1.263V.
