Intel's Prescott may be monumental step back for chipkind

PRESCOTT IS ONLY days away, and it is being greeted with bated* breath and collective yawns, sometimes even in the same review. Most places all say the same thing, it is marginally faster than its predecessor, if that. The differences come in the interpretation of those numbers. In other words, is this marginal gain/loss the next big thing, or a monumental step backwards.

Let's start out with what Intel has said about the chip, because basing expectations on fanboy site ravings isn't really fair. Intel, as far as non-NDA briefings have gone, has said just about nothing regarding Prescott performance. It has categorically not said it will be the next big thing, and more specifically has not said it will beat Northwood.

If you are wondering why it hasn't been shouting the performance from the rooftops, it is because it knows that there is little if any performance gain on the new chip. Simply put, it uses more power, equals the performance, and clocks no higher. But it is very late. Nothing you would want to brag about. The things it wants to shout about won't come out till the socket 775 chips hit, and maybe not even then. There is a lot of hidden bits in this chip.

How average is the Prescott's performance? Stunningly so. Using Intel's own numbers, in one of its recent briefings seen by the INQ, it shows four benchmarks, Quake 3, Sysmark 2002, and the Spec twins, FP and Int. When the graphs flash up, they show a 2, 2.2, 2.25 and 2.5x improvement!

Then you look closely at the numbers, and the 3.40E is compared to a 1.50GHz Willamette core. Sure, Gartner analysts and Newsweek editors will espouse the incredible speedup of the new chip over one introduced on November 20, 2000, but come on guys, you can do better than this.

Actually, it does. A little later on in the presentation, it has a set of numbers curiously labeled ‘backup'. These list six benchmarks across three CPUs of slightly more recent vintage, that vintage being about a week from now. They are the 3.40, 3.40E and the 3.40EE, which if you have been following closely, will realise is a Northwood cored beast.

The benches themselves are Winstone 2003 Content Creation, Sysmark 2002, 3Dmark 03, Quake 3, and the ubiquitous Spec twins. Comparing the 3.4 to the 3.4E, the E wins 2, and loses 2 to the straight out 3.4, and absolutely crushes the older core in Spec. The differences are never more than 3% on the non-spec tests, and average out to a whopping .5% leap to victory by Prescott. On spec, the tables are turned, and the E wins by 12 and 6% in FP and Int respectively. So, in hand picked real world benches, Prescott is up by .5% and in spec, it is up by 9%. Kind of casts an interesting shadow on Spec doesn't it?

The really interesting part about the numbers is the performance of the P4EE 3.4. That chip absolutely stomps the life out of Prescott. The margin of victory is an average 11.25% over the same cored 3.4. The Spec margin is an astounding 21.5% on average.

This points out that cache is very very, maybe even very very very important in these tests. Since the cache on the chips are 512K, 1MB and 2MB, the cache on the P4E has one of two effects. It is either tragically too small of a difference to keep all four of the non-spec benches in cache, or it masks the piss-poor performance of the Prescott core. Since these are hand-picked Intel benchmarks, I will leave it up to readers to draw their own conclusions.

So, that brings us back to what Intel promised, which is not much. From its own numbers, it looks like it will deliver on that promise. When Prescott debuts to collective shrugging of shoulders, please don't blame Intel this time for setting expectations too high because it didn't. If you want to blame the Prescott team for the chip, that is another story, but the PR guys did the right thing, Slide 14 aside. If you want to be really fair, wait for the S775 chips to hit before you blame anyone, remember the performance of the S423 P4? µ

* THIS ORIGINALLY read "baited", which isn't the usual way to do things, but seems to fit, somehow. Ed.