AMD Phenom II X4, the return (page 1: Introduction, the 45 nm)

The Phenom is back and it means business. After a difficult start, due among other things to successive delays resulting from the infamous TLB bug, the Phenom had to position itself in the entry level quad-core range. Things were made difficult for AMD whose production costs due to the 65 nm engraving process have been higher than Intel on its quad-cores. At the beginning of 2009, AMD launched the new Phenom, the Phenom II, that according to the manufacturer offers a 20% hike in performance on its predecessor.

At last, the 45nm

The first change is that AMD has gone from the 65 nm engraving process to 45 nm. AMD is using immersion lithography to achieve this finer engraving, still with SOI type wafers. This technique, co-developped with IBM and which will also be used at 32 nm, allows a reduction in the wavelength of the light beam used in engraving by passing it through an intermediary fluid (pure water).

This allows AMD to engrave more transistors on a reduced surface area, with 758 million transistors on a surface of 258mm² for the Phenom II as against 450 million transistors on a surface of 285mm² previously. Moreover, finer engraving should normally also give lower energy consumption.

What are the extra 308 million transistors used for? Mainly for the cache. Although the L1 and L2 caches are unchanged at 128 KB and 512 KB respectively for each of the four cores, the common L3 cache has been increased from 2 to 6 MB. This is what is bringing about such a gain in performance at equivalent frequencies; a gain of 5% according to AMD.

In this comparative image, not taking account of scale,
you can see that the Phenom II (on the right) has a much larger L3 cache

The manufacturer also attributes a 3% gain to various modifications within the cores themselves, mainly in terms of the prefetching mechanism that anticipates data loading from the memory to the cache.