Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

LOSTCIRCUITS	SHORTCUTS: Top Page Die Shrinks and Leakage Currents Test Setup Amorphium3 Cinebench 2003 I Cinebench 2003 II MDK2 3DMark2001SE Comanche4 Unreal Tournament 2003 X2 Sysmark 2004 Sysmark2004 continued Conclusions Prescott 3.0 Pricing Give Us Some Feedback to Improve our Reviews
Intel Pentium4 "Prescott" To Scale or not to Scale, that is our question.
(Review by MS, May. 23, 2004)

There have been legions of Prescott vs Northwood reviews that were already posted and in so far, it is difficult to come up with really new data. On the other hand, there is always the possibility of putting things into perspective and this review may just have done that.

Amongst the more important issues we were concerned with here was the question what scaling actually means. Overclocking without changing the multiplier will shift the system bottleneck towards some of the peripheral components but it will hardly offer any insights into the effects of architectural differences between CPUs belonging to the same super-family.

In other words, the fundamental issue here is the effect of increasing multiplier settings on performance. That is, each multiplier increase also increases the overhead of the CPU compared to the system bus and introduces additional latency cycles for the execution units until they will receive data and instructions. In theory, this is what SSE or any other SIMD instructions are targeting, that is, a reduction of the data (including instructions) that are needed to carry out the necessary operations.

However, SIMD only temporarily closes the gap, SSE, SSE2 and SSE3 are additional steps that all fall under the same umbrella of providing a temporary remedy for an increasingly worse ratio between the CPU’s execution units and its I/O interface. A larger cache is another solution, in that more data can be held in a superfast on-chip memory that operates at clock speed, that is, independent of the multiplier. In the case of the Prescott, the larger cache is bought at the expense of its access speed as we showed in our original Prescott coverage. Increased latencies are found on the Level2 as well as on the L1 data and trace execution caches and those latencies are amongst the features that partially negate the increase in size.

Back to the topic at hand. Arguably, we have found a few benchmarks that scale better in the case of the Prescott than in the case of Northwood. However, it is necessary to keep in mind that the overall performance in e.g. 3D Rendering applications sees the Prescott at a rather huge handicap to begin with. Therefore, system bottlenecks as constituted e.g. by the system memory do not weigh in at the same level as they might in the Northwood.

All of this leaves office applications as a genre of application where the Prescott really excels and, more importantly performs very well at lower clock speeds and scales to the top of class position at higher operating frequencies. The only remaining question is whether these are applications that really need the extra power.

next page:    => More =>