In One Corner
IBM and AMD are supposed to announce an improved SOI at a chip conference shortly. The Inquirer says that this will result in a substantial speed increase
No doubt we’ll get it straight from the horse’s mouth very shortly, hopefully with more specifics.
There’s still a lot of unanswered questions, mostly in the “when” and “how you can tell them apart from others” categories, but it’s probably safe to say that by the time the seasons flip, we’re probably looking at 3Ghz capable Hammers (which would translate to about 4.5GHz-4.8GHz in PIVese).
In The Other Corner
Intel will start putting out its lineup of 2Mb cache, x86-64-enabled processors starting in February. The maximum speed of those will eventually be 3.8Ghz (3.6Ghz for starters).
These chips will also have Enhanced Speedstep. Given that a more accurate name might be Enhanced Throttling, this is definitely not what the overclocking doctor ordered. Serious overclocking ought to be attempted only by the Mr. Freezes in the audience.
While these chips ought to be fine for your Joe Sixpack Dell box (the EM64T is good insurance against quick obsolescence); they’re certainly going to get beaten by any near-, at-, or above 3GHz “regular” Hammer (and even a 3GHz Sempron will at worst be close).
That’s in 32-bit. If these chips didn’t have x86-64, it could have gotten really ugly in a few side-by-side comparisons. That’s why Intel changed its tune on x86-64; they can keep it close against AMD with it despite their broken-down ramp. They can’t without it.
Nonetheless, the only reasons why any overclocker interested in performance wouldn’t go AMD in 2005 would be either blind hatred of the company or (a bit more rationally), the hope of eventually sticking in a dual-core tomorrow into an LGA775 mobo bought today (which I don’t think is the best idea in the world for an overclocker; if cooling one Prescott is too much, what do you think cooling two near-Prescotts on a single die is going to be like?)
Early Word on Dualies
The Inquirer article also says that Intel ought to have word on dual-core pricing in February. It will probably be hard not to include speeds with that, but we’ll probably see dual-cores running in a range of around 3Ghz each.
According to Intel’s estimates, they’ll get dual cores out the door in 2005, but they won’t become big-time until 2006, when they expect 70% of desktop CPUs to be dualies.
Effectively, that means dualies won’t go mainstream until 65nm, and frankly, the big question on dual cores is not about hardware, but whether the software folks will make something out of it, and how soon.