In AMD's entire history as a mainstream microprocessor manufacturer, they have never seen as poor execution as they did with the K5. After several months of delays, the chip finally hit the streets only to be a CPU that ran too hot and couldn't offer very competitive performance.

On the flip side, AMD had never seen their engineers on top of their game like they did during the days of the K7. What eventually became the Athlon microprocessor was truly a feather in AMD's cap; what ended up holding back the success of the K7 core was sheer marketing and politics - two things that even the best design teams out of AMD could not dream to control. Even with the disadvantages AMD was faced with, the K7 still went on to become the underdog's best effort to date - and thus when they announced its successor, the K8, at the Microprocessor Forum in 2001, expectations were very high.

We concluded our first article on AMD's K8 (aka Hammer) architecture in October of 2001 with the following:

AMD is clearly not the company it was a few years ago. They are constantly making steps towards becoming more of an industry leader as opposed to the follower they have been criticized of being for so long; the Hammer architecture is the most vivid depiction of what sort of an industry leader AMD is capable of being.

At the same time we shouldn't discount Intel as they still hold the majority of the market and they do have the potential to take their technology very far. What AMD's recent gains do prove however is that there won't be a return to domination for Intel anytime soon; this two man race will be continuing for some time to come. Both AMD and Intel have had their slipups; while Intel's have been more recently, AMD is far from immune to them.

The technology behind Hammer is there, as is the potential for it to succeed. But AMD has a lot of work to do between now and its release in the next 12 months. Many forget that until the Athlon, AMD didn't have the best execution track record. It's a long road ahead for the Hammer design team, good luck guys.

Fast forwarding to today, it's clear that AMD not only didn't hit their target release with Hammer, but the process was far from another K7 launch. We asked AMD's CTO, Fred Weber, to characterize the execution of the K8 launch and he fairly summarized it as not as good as the K7 but better than the K5/K6.

It has been a very long road for AMD; dealing with a new microprocessor architecture, never-before-used manufacturing process (Silicon on Insulator) and all while facing very tough competition from Intel and a weakening world economy. The light at the end of the tunnel is finally able to be seen, and although the K8 isn't going to be available on desktops for another several months, its workstation/server release is here and it's called Opteron.

This article will focus exclusively on the architecture of the K8 core and how it relates to Opteron, we have three other pieces published in tandem with this one that cover everything from enterprise performance to desktop performance of Opteron. Here are direct links to the other articles:

AMD's Opteron Coverage - Part 2: Enterprise Performance
AMD's Opteron Coverage - Part 3: 1U Server Roundup
AMD's Opteron Coverage - Part 4: Desktop Performance

We strongly suggest reading this article in its entirety before proceeding onto the rest, unless you're already intimately familiar with the K8 microarchitecture.

With that said, and without further ado, let's get to it.

Designing a CPU, More Specifically the K8
Comments Locked


View All Comments

Log in

Don't have an account? Sign up now