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 Strained Silicon Die (re)Organization Prescott New Features I Prescott New Features II Test Setups Latencies Sysmark 2004 Internet Content Creation Office Productivity 3ds max 5.1 Amorphium Cinebench 2003 D3D Gaming => OpenGL Thermal Issues Overclocking, Final Words Prescott 3.0 Pricing Give Us Some Feedback to Improve our Reviews
Intel Pentium4 "Prescott" Strained to the Silicon
(Review by MS, Feb. 1, 2004)

In absence of higher multipliers we had to resort to bus overclocking. At the default Vre of ~ 1.4V and a 16 x multiplier, we were able to run the Prescott sample at 230 MHz bus speed. Increasing Vre to 1.55 V allowed us to run up to 240 MHz bus speed, which translates into 3.84 GHz. That is, at least, we managed to get into the desktop and to run a few benchmarks, however, the system was not entirely stable at that speed. To exclude the memory as culprit, we were running the memory at the DDR320 setting in asynchronous mode. In other words, we have run the same system before at higher bus speed without crash but also at lower multiplier settings (e.g. 12 x 300 MHz).

With respect to performance scaling, we really cannot make any statements since the overclocked system bus would skew the results and adding an asynchronous memory mode would make things even worse.

Getting up to 3.84GHz using standard cooling appears rather reasonable, however, with the Northwood sample we have at hand, we managed to get 3.66 GHz under the same conditions and the resulting delta of 180 MHz is not spectacular enough to warrant the transition to another processor.

Conclusion

Intel is releasing six new processors, that is, aside from the four (counting the 3.4 GHz version) speedgrades of Prescott, there are the 3.4 GHz versions of Northwood and the ExtremeEdition. Most of the argumentation in favor of Prescott has focused on the scalability issue, that is, higher performance gains at higher clock speed and, by extension, higher multiplier settings.

What this comes down to is the issue of latencies and the mismatch between the processor and system bus and the actual core speed. The most effective way to mask this discrepancy is to force data access from the CPU-internal cache rather than having to resort to outside sources - e.g. the system memory. This is where the doubling of both Level 1 and Level 2 caches will make the biggest difference in performance.

On the other hand, Prescott is looking at some massive increases in latency, the access latency for the Level 1 cache has quadrupled, and the Level 2 cache accesses are approximately 50% slower. This appears where the scaling goes the opposite way with lower clock speed. That is, lower clock speed and higher latencies will cause a rather severe performance penalty and we had a few examples of that in our benchmark section.

One needs to keep in mind though that the current samples of Prescott are the beginning of a new technology and there may still be some raw edges somewhere in the process. Power consumption appears to be relatively high, to say it mildly, heat dissipation is way above that of either Northwood or ExtremeEdition. However, we have seen similar issues with the first stepping of the Thoroughbred core that, after adding an additional metal layer and a few decoupling capacitors turned out much more civilized, in other words, we better wait for the next revision before making definite statements.

On the up-side, we have a new record in transistor gate length. Compared to the 60 nm gates used in the Northwood, the shrink to a 45 nm gate in Prescott scales closely with the reduction in interconnect technology, namely from a 130nm to a 90 nm process, so everything is approximately in the same league. At those dimensions, that is 45 nm with a 1.2 nm gate oxide thickness compared to the former 60 / 1.5 nm (estimated), we can expect future revisions of Prescott to achieve some 4.5 GHz clock speed. This is with the caveat that Intel gets the power consumption and thermal dissipation under control. Otherwise, we will be looking at some 115 W power consumption and thermal dissipation for the high-end models.

next page:    => Future Outlook and Final Words =>