When Intel launched its new .09 micron Prescott chip back in January, it took the unusual step of sampling only 3.2 GHz chips, while announcing that the 3.4 GHz CPU?s were launched, but with availability to follow.  While this may have been more honest than sampling the media with parts that the mainstream market wouldn?t see for weeks, its still not exactly a launch, which is why we refused to call it such.  Instead, we stated that Intel was ?announcing? the 3.4E, with availability to follow in a few weeks.

Six weeks later here we are, and Prescott 3.4E?s are now beginning to trickle into the market.  At the same time, however, there?s not a whole lot about Prescott that makes it tremendously appealing.  It?s already been demonstrated to be slower, on the whole, than a Northwood at an equivalent speed, and it?s almost guaranteed to be slower than the P4 3.4 Extreme Edition (which also has a 3.4 GHz component).  It?s also currently running so much hotter than a 3.4 GHz Northwood or even 3.4 EE, that we wouldn?t recommend one for that reason alone.

We?ve already seen the apex of 3.4 GHz performance in the EE review six weeks ago.  Rather than re-testing Prescott 3.4 at 3.4 GHz and handing off another set of benchmarks which would demonstrate the CPU losing some by whopping percentages, tying most, and winning a bare handful.  Instead of whipping up another review to cover performance that?s already been exhaustively discussed, we decided to look at Prescott from a different perspective, and investigate its scaling.

The Point of Scaling Prescott:

One of the claims Intel has made about Prescott is that the CPU would scale more effectively than Northwood over time.  This is an intriguing claim, if true, and would explain why Prescott may become a more attractive purchase over time.  Initially, however, it was difficult to see much of a difference.  Moving from 2800 MHz to 3200 MHz is a difference of only 14%.  Allowing for a margin of error in benchmarking made it almost impossible to see any marked difference between CPU?s.  For this reason, we?ve overclocked our Prescott sample to 3.8 GHz, on a 224 (896 MHz) bus.  AGP and PCI busses were locked.  RAM speed was kept 1:1.  This is not meant to be a direct representation of what 3.8 GHz Prescott performance will be, but by putting both it and Northwood on an equivalent platform, we should be able to tell if Prescott does, indeed, scale better.  At 3.8 GHz, Prescott is running 36% faster than at 2.8 GHz.  This is a large enough gap to see a difference, and if the scaling advantage takes longer than 3.8 GHz to manifest itself its doubtful consumers would be all that interested.  Intel, after all, is forecasting only up to 4 GHz this year. 

If Prescott truly does scale better than Northwood, than we should see the percentage gap between the two CPU?s grow over time.  In tests Prescott loses at 2.8 GHz, it should lose by a significantly smaller percentage at 3.8 GHz, while tests that it wins at 2.8 GHz should be won by a larger amount at 3.8 GHz.  Since different programs scale differently we?ve investigated a number of programs in our benchmarking suite (though unfortunately we didn?t have time to do them all).

We do, however, want to spend some time discussing Prescott 3.4E before we move on to other factors. 

next >

---

• Prescott at 3.4 GHz:  Nothing Different Yet (1 of 6)
• System Configuration (2 of 6)
• Prescott vs. Northwood: Game Scaling (3 of 6)
• Prescott vs. Northwood: 3D Rendering (4 of 6)
• Prescott vs. Northwood: Synthetic Tests (5 of 6)
• Analysis, Conclusion (6 of 6)