AMD Phenom 9600 (page 1: Introduction)

Launched in September 2003, the Athlon 64 was a true success. AMD had an architecture that was extremely efficient compared to its rival, Intel, whether it was in terms of brute performances or performance / power consumption ratios, an aspect which started to take on increasing importance. AMD plowed ahead with the Athlon 64 X2, or its dual core version, which arrived in May 2005.

Intel’s reaction was slow as it took 3 years for the Core 2 to arrive last summer. However, the response was worth the wait because Intel has since offered an architecture that doesn’t suffer from any real defects and produces first rate performances despite its reduced power consumption. The Santa Clara giant drove its point home even more by launching the first x86 quad core processor, the QX6700, a year ago.

One year later, AMD finally responds to Intel’s quad core with the Phenom, which is based on K10 architecture and to which we devoted an entire article. Of course, this was longer than officially planned as last year mid-2008 was the target and even six months ago it was a question of the third quarter.

A « native » quad core

With 463 million transistors, the AMD Phenom processor is comprised of 4 cores. Each is equipped with 128 KB of L1 cache (as was the case with the Athlon 64) and 512 KB of L2 cache. In addition, the cores have a third level cache with a capacity of 2 MB.

AMD underlines the « native » quad core characteristic of its processors as opposed to Intel. And there are indeed a number of characteristics which involve the ensemble of all four cores together instead of groups of two like with the Core 2 Quad. This is especially the case for L3 cache and the fact that the Phenom is composed of a single silicon die unlike Intel’s solutions.

AMD and Intel were already were more or less in the same position at the launch of the first dual core processors and actually the advantage of the « native » aspect wasn’t too evident on desktop configurations. Gains between 1 and 2 cores were very close on the two different architectures at the time.

On the other hand, the management of the defective products is quite different. Intel assembles two dies of 143mm² (65nm) or 107mm² (45nm) to produce a quad core, which means that in the case of a defective die only half of a quad core is discarded. For AMD, a K10 is 285mm² and when possible they can recycle partially defective processors into tri and dual core versions.