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: A new bus speed Heat and Hertz Test Configurations Cachemem, Sandra Everest and Speed Derating Sysnark2004 Internet ContentCreation Office Productivity WorldBench5 -1 WorldBench5 -2 3D Rendering 3ds max Cinebench2003 Futuremarks Gaming A-Z DOOM3 FarCry Conclusions and Paradoxes Give Us Some Feedback to Help Us Improve our Reviews
Intel Pentium4 3.46 Extreme Edition / 925XE chipset
(Review by MS November 16)

MegaHeat Mega-Hurts MegaHertz

AT LEAST by now, it is common knowledge that the latest versions of the P4 suffer from rather intense thermal dissipation syndrome. Part of this originates in the progress in interconnect process that, at 90 nm design specs, no longer warrants efficient insulation of the traces. Despite advances in transistor design using compression and stretched silicon straining techniques to improve the electron flow across the gate, it has not been possible to lower the operating voltages to a level where leakage currents are no longer a predominant factor and severely impact thermal and power management.

Power management countermeasures have involved highly sophisticated voltage regulator module solutions in multiphase configuration. The realm of thermal management, on the other hand has lately just pushed the envelope of physics with evolutionary improvements at best. To be fair, the entire field of cooling is one of the areas of applied physics where, off-hand, the frontiers of thermal management appeared to be reached already some 5 years ago. Yet, looking back, especially in view of the progress made by detail improvements, those were still the dark ages compared to what has been accomplished lately.

         

Cooling from above and beyond: the massive tower takes care of most of the heat and blows a bit of air through the VRM cooler, the memory is using heatspreaders and, in addition, the "Cool Stack" on the back of the mainboard will supposedly carry away a bit of the thermal load as well. Considering the dead air space behind the board, this does not appear a very efficient solution -- but-- whatever helps.

The original Pentium had a metal plate covering the substrate of the die and that solution sufficed. The Klamath (.35µm) and Deschutes P2 along with the Katmai core-based P3 incorporated some more elaborate cooling solutions. The Coppermine introduced the flip-chip design in order to move the source of heat to the surface. The Pentium4 finally introduced the heat-slug, a full metal jacket à la Stanley Kubrick, originally conductively attached to the core by silicon grease with zinc-oxide for better transfer. Throughout the various iterations of the P4, also the thermal interface material changed, finally reaching a low melting point solder to permanently bond the core to the slug.

The result is better dissipation of heat and faster cooling off of the core, especially after high transients, as sinks we have the rather monstrous bimetal coolers. Internally on the CPU other heat management measures are in place, the most important one is the thermal throttling, which, at last count includes 6 different mechanisms that were added one by one to the different versions of Intel CPU and never taken out again. Practical use of the thermal control circuitry (TCC) and the thermal diode is somewhat nebulous and not only varies between CPU steppings but also between the different mainboard designs and the feedback circuitry used to talk to the machine specific registers (MSRs) on the CPU.

Fixed, however, is the factory-calibrated thermal throttling point that is usually set to 70 degree Celsius at the hottest spot of the die – wherever that may be, which implies the presence of multiple sensors that each can trigger throttling. Suffice it to say that at any elevated average die temperature, a local hotspot can exceed the throttling threshold before the heat is carried out to the heatsink. We will talk about the relevance of this a bit later in this article in the context of memory benchmarks.

For the time being, suffice it to say that the latest trends have entailed a migration to larger caches or, in the case of the Extreme Edition, towards an additional level of cache. The latest and most significant step, though, is the move from an 800 MHz data rate PSB to a 1066 MHz host bus interface. Interestingly, as we will show a bit later, the Northwood – Galatin core based Extreme Edition cannot take advantage of the increased bandwidth as well as the Prescott stepping with its deeper pipelines can. Let the benchmarks begin….

next page:     => Test Platforms =>