The Pentuim4 Willamette Platform

The P4 Willamette processor will be introduced later this year in the $2000 and higher range, matched with its 850/Tehama+RDRAM platform. Today's ICH2 will be used. The risks associated with this platform are known. Tehama is essentially the same as today's 840 / AGP4x chip set with a new front side bus interface for the P4. The Tehama MCH chip is depicted in the photograph below, using flip chip package technology and apparently requiring a heat sink.

Willamette is rumored to be able to accommodate front side bus speeds of 100 and 133 MHz as indicated in the Tehama board photograph above (first published by the Register). It is also rumored to be able to support 2x and 4x data rate modes for the FSB. These combinations would enable FSB bandwidths of 1.6GB/s, 2.1, 3.2 and 4.26 GB/s. Tehama is likely to make use of the processor's 100x4 mode exclusively, matched well to the dual channel RDRAM subsystem.

Though this extremely high bandwidth front side bus would be particularly useful in dual processor configurations, Willamette will only work in single processor mode. Dual processor options will be restricted to the Foster implementation of this CPU aimed at the server market, with large L3 caches and much higher price tags.

Primarily because of its deep, 20 stage instruction decode pipeline, the Willamette processor is expected to deliver less than exciting performance in mainstream applications, as compared to P3 or Athlon at a given clock speed. Assuming identical clock speeds, the P4 is estimated to underperform the P3 or Athlon by more than 20%.

On the flip side, the primary advantage of its 20-stage pipeline is greater clock speed scalability. Willamette will be able to reach higher clock speeds than the P3 in the same manufacturing process (0.18 micron in this case). We are expecting speeds of 1.4 or 1.5GHz at launch in Q4'00, but we do not expect these systems to decisively outperform 1.2GHz Athlons or P3s that should be available in the same timeframe.

Since the release of Intel's performance analysis of the 815+PC133 vs. 820+RDRAM, one might expect RDRAM to share in the blame for Willamette's potentially unexciting performance profile. But other factors may also be brought to light, such as complier optimization, and more complaints about inadequate benchmarks.

Prev | Next