Intel will reveal details of its 65nm generation of processors next week.
They’ll have all kinds of names for them, but these processors will be the love children of Grandson of PIII (aka Dothan) and Daughter of PIV (aka Prescott), and inquiring minds will want to know if the kids look more like Daddy or Mommy.
AMD is also supposed to be skulking around the Intel conference giving details on what it plans next, too.
The Single Question For Overclockers
There is one, and only question overclockers need answered about these chips, whether from Intel or AMD:
“Can we expect what used to be the normal increase in computer power from the 90nm-65nm shrink, as in 40% or more increase in single CPU power?”
In other words, are we going to be able to get back to the good old days?
From all advance indications, the answer is “No.” It looks like something in the range of 15-20% is more like it.
Of course, neither party is going to flat out say, “We can’t deliver like we used to.” However, the more they talk about multicores being the only way to go, the more they’re really saying, “We can’t deliver like we used to.”
The problem with multicore for speed freaks is that multicore is only good for speed freaks when it gets enabled by software, so software releases will become more important than hardware releases.
As we mentioned more than once in the past, multithreading doesn’t benefit everything a lot, and sometimes can’t benefit some things, even if you write code from scratch.
Not to say multicore doesn’t have other legitimate benefits for other users, but for speed freaks, the benefits at best will be delayed and often won’t ever show up.
The Real Problem And Solution
The real problem any CPU manufacturer faces today is that circuits leak too much power when running at high speeds. For Intel, it became a big problem when the PIV was introduced, and became overwhelming with Prescott.
IBM/AMD adopted SOI to fend off the problem, but found that SOI was a bear to implement, didn’t help quite as much as expected, and had its own problems when pushed hard. It left them in better shape than Intel, but they just have a milder fever at the cutting edge.
All the CPU manufacturers have been desperately searching for ways to fix this problem, and they haven’t found an easy fix that they could implement into the 65nm generation of products.
So Intel is buying time by throwing its notebook chip into the picture for a couple years. AMD doesn’t have a magical formula, either, so they’ll tweak themselves a bit more along. Both will add an extra core or three and hope people don’t notice a lot of things really aren’t running a lot faster.
It’s not that the CPU manufacturers haven’t found any solutions at all; it’s just that they aren’t easy to work out the bugs/implement. They involve big changes in both materials and CPU circuit geometry, so it isn’t surprising we won’t see them next year.
We have seen that going to SOI has given both IBM and AMD huge headaches over the past few years, and SOI is a trivial change compared to those being discussed today. These are real, serious legitimate problems, and when the solutions cost gigadollars to implement, it’s understandable that these companies want to keep look more before they leap.
Unless there’s some real surprises next week, we’ll probably have to wait until 45nm until we see the best of the lab successes in the next shrink.
That will mean we can’t expect much out of 65nm, and any choices you make will not be mostly performance-driven.
Be the first to comment