Athlon64 3200+ 0.09µ Socket 939
“The Next 2.4C?”Introduction:
Since the dawn of time men have sought to invent things and consequently improve upon those formulations. Women just mainly get mad at us that we do not get it right in the first place, but I digress! When it comes to improving computer hardware however, there is no stalemate or truce among rival companies. To do so would be tantamount to failure, just ask 3DFX, creators of the much loved Voodoo 3d Graphics. CPUs (Central Processing Units) or the “brains” of our computers are in constant flux when it comes to development. “Smaller, faster, better,” are the words I imagine being yelled out of many a boardroom when engineers meet with the corporate bean counters.
Nevertheless, while smaller, faster, and better are ideals that are strived for, making something smaller and faster doesn’t necessarily equate to a better product. The Pentium 4 Prescott CPU has suffered a somewhat awkward stage during the several months surrounding its launch. Sure, it is built upon the 0.09µ (micron also written as 90nm) die process, which in theory should cut heat output and power requirements. History though has a way or rearing its ugly head, like it was the case with the initial 0.13u (micron, 130mn) transition of AMD’s AthlonXP (thoroughbred core), the Prescott was plagued with heat problems that were apparently attributable in part to leaking transistor gates.
AMD eventually came up with a workaround for their 0.13µ Thoroughbred parts and the AXP became successful in that process and eventually spawned the Barton core with added L2 cache. Intel seems to be getting a handle on the heat problem with the Prescott cores and later steppings are reportedly much cooler running. In addition, the Prescott was supposedly given core enhancements, additional SSE instructions, and a whopping 1mb of L2 cache. People were wildly speculating the demise of AMD due to this CPUs projected performance and paper statistics. The Prescott at launch ended up proving slower clock for clock than its sibling, the P4 "C" (Northwood core) CPU in a majority of applications, thus showing smaller and faster didn’t necessarily equate to better, and certainly not cooler.
So one might ask, aside from the obvious headaches in transitioning a part to a smaller die process, what are the advantages of a process shrink. Well as a general rule, heat output is decreased due to lower power requirements that usually are a byproduct of smaller transistor gate. This again has not always been the case though, as pointed out in the above situations. A smaller refined process, if implemented properly can lower heat output and thus allow engineers to scale the clock speeds of these chips higher as a result.
The average consumer is oblivious to such nuances of CPU architecture and improvements, but then again I doubt any of our readers are the average consumer :-). Finally, another advantage of a smaller die process is that more CPU cores can be produced on a single wafer, increasing yield (theoretically) and consequently lowering costs. Eventually that cost saving spills over to the consumer in lower CPU prices, something I think all of us can agree on; lower prices equals a happy consumer!
This then brings us to the focus of this article. AMD has recently begun the process (pun intended) of producing and marketing CPUs based on their execution of a 0.09u (micron) process shrink. Code-named Winchester, the die shrink for the Athlon64 is initially for the Mobile Athlon64 (socket 754) and socket 939 Athlon64’s (non FX series). Supposedly, core enhancements during the die shrink are reported to bring a small increase in performance over 0.13u New Castle core processors. Eventually, AMD’s whole product line will undergo changeover to the newer/smaller process. Let's find out if those claims are true ->