It’s been long known that Intel’s new CPU strategy is called “tick-tock.”
Tick-tock is essentially “shrink the process technology of the previous architecture (tick), then use that process technology for the next architecture (tock).
This raises the issue, “If you can’t buy both the tick and the tock, which one should you buy?”
The answer to that question for overclockers depends on what change is provide the most improvement.
If a process shrink is likely to provide a big improvement through increased speed, the “tick” is a better choice. On the other hand, if process shrinks yield little improvement, it is better to buy the “tock.”
If this were the year 2000, it would be hard to argue against the “tick.” Back then, a simple process shrink yielded large performance benefits.
Those days are over. Penryn demonstrates the level of improvement you can expect from “tick,” not a whole lot.
That’s not a knock on Intel. This is an era where the priorities are more cores while staying within power envelopes, using new, exotic materials just to make it work.
Given that, a 10-15% increase in maximum potential speed is probably more of an accomplishment than getting 50-75% faster during the dot.com era. If you think not, just how has AMD fared with 65nm?
It’s just not very thrilling.
In comparison, the “tock” (in Intel’s case, Nehalem) offers a new CPU architecture, like C2D (actually, Intel’s first “tock.”). If nothing else, it will have an integrated memory controller, which was the primary reason for the jump in performance between AMD’s K7 and K8.
(OK, the “tock” isn’t going to get you anything like 50-75% improvement, either, but it will probably be better than the “tick.” If you’d like further proof, consider the 65nm PIV as the first “tick” and C2D as the first “tock.”)
Another advantage of the “tock” is that it’s more likely you’ll be able to upgrade to the next “tick,” should you deem that wise.
Hardware and software features of the “tock” platform are also more likely to be used right away, or at least be mature and affordable. You’ll pay a good deal less for a ton of faster DDR3 memory in a Nehalem system than you will for a Penryn system.
Even features first included in the “tick” (or before) are more likely to be actually used for the “tock.” For instance, SSE4 is Penryn’s best performance feature, but it probably won’t be used entensively until Nehalem comes out. Effective multithreading ought be more commonplace by the time Nehalem comes out, and maybe, just maybe, after Vista SP1, we’ll see some 64-bit applications show up.
What about AMD, you might ask?
AMD’s problems can be largely described as a matter of not ticking and tocking right.
K8 was a “tock.” It was such a big “tock” than AMD didn’t think they had to seriously “tock” again for another decade.
(To be fair, Intel wasn’t much better at tocking, This was why the PIII and PIV eventually turned into ticking timebombs for Intel.)
K8 was a fine design but a botched process “tock.” AMD couldn’t build them right until the 90nm “tick.” That was fine, but then AMD couldn’t get the 65nm “tick” right, and in any event, they had no “tock” to match that of Intel’s C2D . K10 is supposed to be that “tock,” but it certainly isn’t a “tock” like C2D, and it still is suffering from the problems caused by the 65nm tick.
45nm K10s will be just that, a “tick.” We won’t see another “tock” from AMD until the K11 generation sometime in 2009.
Whether it’s Intel or AMD, the key reason why the “tock” looks better than the “tick” is that process shrinks just aren’t going to provide much of a performance advantage any longer. In all likelihood, future architectural changes for both Intel and AMD will be rather less profound than in the past, more than tweaks, less than a revolution, but that will beat hardly any improvement at all.
And no, none of these reasons are watertight, but taken all together, they’re definitely biased against the tick and for the tock. Once you’ve adjusted your own buying clocks, you ought to be, too.