Rounding the Dual Core (Wood) Crest
While Yonah was truly the first ground up dual core design, featuring a shared L2 cache, some of the benefits were lost on a mobile platform. The true benefit of a good Chip Multi-Processing (CMP) design is that it significantly reduces cache coherency snooping, which is a major issue for workstations and servers, but less so for mobile and desktop devices. In that regard, Woodcrest will substantially improve over prior server chips, which as noted in an earlier article, do not share any cache. Furthermore, Intel has hinted that Merom can transfer data directly between the two L1D caches. The architects declined to disclose any further details on this feature, other than to say that it exists. Figure 1 below shows a high level system comparison between Yonah, Merom, and Dempsey.
Figure 1 – Architectural Comparison of Intel MPUs
From a high level perspective, Merom has inherited the CMP hierarchy of Yonah, using a shared last level cache and bus interface. At the same time, the bandwidth for Merom is closer to that of Dempsey than Yonah; the on-chip caches run substantially faster and the front-side bus runs at twice the bandwidth. As noted above, these improvements help multi-socket systems more than desktop and mobile machines.
Intel will release Woodcrest products first, in early Q3, followed by Conroe later in Q3 and Merom in Q4 (or possibly Q1 of ’07). These release dates are more of a reflection of Intel’s competitive position in each market and their margins, than anything else. The mobile market has the highest margins for Intel, although the least competition, thanks to Intel’s aggressive marketing of the excellent Centrino platform using Yonah. This is not expected to change in the near future, and Yonah was released very recently in January. The server market has the next highest margins, and Intel’s offerings are lagging behind AMD by a fair bit. The desktop market is the least lucrative of all three, and is moderately close in terms of performance. Looking at all these variables, it is easy to see how and why Intel picked their release schedules. The server offerings need the most improvement, then the desktop market and last the mobile market. Moreover, a short product life cycle for Yonah would be financially and strategically undesirable.