Looking backward, looking forward...
It is the time of year for such reflections, after all.
When AMD rolled out the Opteron/K8 family, they made a further commitment to the balanced IPC/clock speed design choice, which was technically very well justified. Intel saw a chance to regain marketshare by emphasizing megahertz, to the consternation of their engineers. From a marketing psychodynamics perspective, it made sense, given that Joe six-pack saw the MHz he was buying, and was wholly unaware of IPC. We all know where this approach ended up, but note that a major new processor family such as MCW didn't just spring up like mushrooms after Prescott hit the thermal wall. For much of the time that Intel was shouting "Look over here, everybody!", they were slaving away on a balanced IPC/clock speed response to the leadership path that AMD had established.
The new "Look over here, everybody!" is core count. 4 GHz, four core; 10 GHz, 16 core. Intel has seized upon core count as the new MHz, as witnessed by their rush to multi-die assemblies and the hoopla over quad core desktops. AMD has recognized the true strength of quad core, and is reading their single die, ODMC and cHHT empowered Barcelona for the server market, where it will return the crown to AMD. They are also planning an octal core server processor, because the inherently multithreaded nature of server loads can readily exploit eight cores, and the server market happens to be a very lucrative one. Intel once had the x86 server market all to itself, and AMD's strong arrival here has caused a painful hemorrhage of Intel's OP. High core counts are all wrong though, for the vast majority of the CPU marketplace, due to the simple fact that the vast majority of desktop and mobile software is and will remain single threaded. So why the fuss? Intel is currently at an architectural and graphics sophistication disadvantage to exploit anything other than core count, which is their "new MHz". Heterogeneous cores, accelerators, fast on die interconnects, and graphics sophistication are the key to the next advances in both non-server and server processors, and AMD is leading this charge. Phil Hestor's slide "No core wars" is right on the money (literally). Buying Nvidia would help, but sorry, HSR says "No deal". We can expect Joe six-pack to be bombarded with "But we have more cores!", but the real question is: What is Intel working on now for when that excitement proves hollow? Just as MCW didn't spring up out of nowhere, don't remain focused on the magician's hand. The switch is being made elsewhere. As before, though, AMD gets there first and shows the way forward. In a word, leadership. |
