I focus on the past because one data point doesn't make a trend. If this node and the next node both offer good gains in frequency when shrinking existing designs, I'd be inclined to say the trend has changed. Basing your prediction of a major improvement on a single data point when the last few before it didn't pan out the same way is being hopeful, not logical. Also keep in mind that I'm only talking about shrinks of an existing architecture and that I'm only commenting on node performance at introduction. AMD's processes always mature nicely. They just usually suffer a rocky start.
Bookmarks