I got a note yesterday from a long time reader who asked two questions, one nuts-and-bolts, the other more philosophical.
The nuts-and-bolts question was, “Why the pessimism about Conroe overclocking?” He pointed out the rumors of a “regular” 3GHz and a 3.33GHz Extreme
Extortion Edition Conroe.
Fair enough question, and the answer to that can be found in these Conroe roadmaps
First, the fastest “regular” Conroe, the E6700, will only run at 2.66GHz. This by itself means little, but the roadmap shows the E6700 as top dog through Q1 2007, no newer, faster one scheduled.
If Intel could make them run faster than that, they would. They aren’t, so odds are, they can’t.
Second, while a Conroe XE is listed in the roadmap starting in the third quarter, there’s no technical information about it, unlike the other Conroes. This tells me that they haven’t gotten a final speed down yet on it.
XEs have generally not run a whole lot faster than the “regular” chips. Usually, they’ve run around 10% faster than the fastest regular chip. A 3.33Ghz XE is believable when there’s a 3GHz Conroe around. It isn’t (unless it really is much more extreme than Extremes have been) when the top regular chip is only 2.66Ghz.
(Yes, XBit Labs had an article about a month ago saying something different. It isn’t clear which roadmap is more recent, so I’ll go with the more conservative, better documented one for the reasons mentioned above, though I could be wrong.
All this leads me to believe that Intel had to scale back their expectation a bit, and the first XE is likely to run at around 3GHz, maybe a little less than that.)
The more philosophical question boiled down to “What makes a CPU a good overclocker?”
There seems to be two schools of thought on that. The predominant school seems to be what I call “the percentage school.” The percentage school judges a chip by the percentage by which it overclocks.
The other school of thought is what I call the “How far for how much?” school. It doesn’t care about percentages as much as the end result.
To give an example, if you bought an Opteron 144 at a scarcity price of $175, and got it from 1.8 to 2.8GHz, the percentage people will say “Wow, that’s a 55% overclock!” They would be less impressed by someone who bought an Opteron 148 for $215, and got it from 2.2GHz to 3.1GHz.
The “How far for how much” school would prefer the Opteron 148 overclock, simply because 3.1 is more than 2.8, and the price gap between the 144 and 148 wasn’t too much. If the gap had been a few hundred rather than a few dozen dollars, that would be a different matter.
With Conroe, what we are likely to see are people flocking to the 1.86GHz chip, looking for a big percentage overclock. As we pointed out yesterday, running the 400MHz+ FSBs necessary for a 50%+ may prove to be a real problem and bottleneck, at least initially.
For sheer performance (and especially if Conroes are capable of more than I suspect), one might be better off with an E6600. It comes with an extra 2Mb cache, which never hurts, and a 3GHz speed would be reached at just 333GHz FSB (compared to a 430MHz FSB needed to reach 3Ghz with an E6300, which I would not count on from a mobo) . That’s only a 25% overclock, but that gives you a faster chip (along with an extra $100 on the bill).
Let’s say Conroes are a bit better than I think, and the initial generation of mobos are generally capable of a stable 400MHz FSB. That would get you about 2.8GHz from the E6300, but 3.6GHz with the E6600.
Let’s assume Conroes are only capable of 3.2GHz, so our E6600 maxes out at a little over 350MHz. Which is the better overclocker, the 50% overclock getting you 2.8, or the 33% overclock getting you 3.2 for an extra $100?
I would say the latter, but many would disagree with me.
No matter what your definition of a good overclocking CPU is, though, all will agree to be a good overclocker, you have to be able to buy one, so we can wait a bit before judging these ones. 🙂