There are initial benchmarks of a 2.8GHz Prescott starting here.
We’ll note on the preceding page that there’s a fairly harrowing temperature, though it’s still pretty warm in Taiwan this time of year.
A few of the benchmarking programs clearly had problems with the Prescott. If the score is well below or above that of a 2.8GHz Northwood, best to ignore it.
Most benchmarks show the Prescott doing better than the Northwood at the same speed, but the improvement can almost always be measured in one-digit percentage points.
This is not surprising. The only immediate advantage a stock Prescott has over a stock Northwood is a 1Mb rather than 512K L2 cache.
Cache is one of those improvements where you have diminishing returns. You can certainly have too little cache (see the Celeron), but unless a particular program can take particular advantage of more cache, it doesn’t help a whole lot.
Overall, Prescott’s bigger cache might get you 5% more performance, and even that figure may well be on the high side.
The reason why Prescott ought to be more appealing to overclockers than Northwood is not because of any particular chip features. It’s because it will be made using a 90nm manufacturing process, while Northwood is a 130nm chip.
Historically, chips made with a smaller-sized manufacturing process can run faster than those using a bigger-sized manufacturing process. That’s what makes Prescott interesting.
However, historically, chips made with a smaller-sized manufacturing process also use much less power at a given speed than those using a bigger-sized one. Current indications are that this relationship has broken down more than a bit with Prescott.
That could easily mean real problems for overclockers down the road, and might mean problems/inconveniences even for those who don’t.
Prescotts will not become affordable until next February, so this doesn’t really affect overclockers until then.
Intel Chips: They Just Grow On You
In the long run, Prescott chips do have another advantage over Northwood: Prescott New Instructions. Essentially, you can call that SSE3. It’s a way to process certain computer functions more quickly and efficiently.
However, it’s up to software to take advantage of such features, and as a general rule, they take their sweet time about it. It normally takes a year or more for software folks to take advantage of it.
This is the principal reason why Intel chips often show jumps in performance compared to AMD chips when new versions of benchmarks using applications come out (granted, sometimes too much so).
These additions tend to help a few programs quite a bit, and many not at all. Most people should consider features like PNI a little extra delayed gravy rather than a reason to buy.
This time around, though, AMD has its own, more potentially potent performance time bomb ticking: x86-64. Only time will tell whether it will go off with a bang, or prove a dud.