Motherboard power circuitry temps overclocking the Prescott 3.0E




The Prescott had finally arrived, and knowing the risks I would be taking even after learning of the chip's defeatist architecture, I went ahead and purchased the world's first CPU built upon a 90nm process, and compatible (at least initially) with Socket-478. I'm no martyr, however; I've done this for you the reader. Knowing Anandtech had tested Prescott up against Northwood, and Northwood bested the 90nm chip in most benchmarks at default speeds, was ancillary to why. This question had to be answered, and furthermore was Prescott salvageable? If not for average desktop computing, would Prescott be an Overclocker's chip? I'd read [H]ardOCP's feature forewarning of the duress Prescott's power requirement would have on motherboard's designed for an entirely different species. My curiosity, however; prevailed, and in anticipation for the 90nm silicon, I purchased the Abit AI7. First I'd like to introduce readers to one of the most comprehensive article's written concerning the engineering path leading to Prescott, Bits and Bytes by Van Smith is a must read.


Now, after extensive testing I can safely say Prescott is not entirely deserving of the reputation which preceded it. Surely the CPU could have been designed better; of course engineering hindsight is always 20/20. Irregardless of its deficiencies, the chip is in fact highly overclockable. My 3.0E SL79L easily attains a steadfast 600MHz overclock (15x 240FSB (1:1)), without any Vcore adjustment. While this doesn't compete with some Northwood's which can run 800-1000MHz beyond default, without Vcore increase, it's still very good. Another issue I can now see that has been exaggerated is Prescott's heat output. While I'm sure my temps are the result of high-quality water-block's, such as the Danger Den RBX, and Cool-Cases CF1, the average Pentium-4 cooler should still suffice for average use. Perhaps the most disconcerting issue where Prescott is concerned is not attributable to Prescott at all.


In searching for a motherboard for my new Philippine 3.0E, it never dawned on me that "Prescott Ready" entailed much more then a prima facie BIOS fix. In fact I was confident "Prescott ready" meant just that. Upon its release, I grabbed the Abit AI7 i865PE based motherboard, which features Abit's new µGuru technology. While µGuru is certainly an interesting feature, quite handy for the budding Enthusiast, it was but one small feature of the board which piqued my interest. For me the AI7 strengths lay in its BIOS, 3.30V DDR, 1.90Vcore, SATA RAID, Ethernet, 5.1sound, and the same Orb aluminium NB/HSF found on the MAX series, all these accoutrements for just $110. Ironically it was one feature included on µGuru utility monitoring GUI which became the inspiration for this article. First off, µGuru derives it's capabilities from a Winbond chip custom made for Abit, seen below;





The utility itself is quite extensive, in that it allows the end-user FSB adjustment from within Windows, harking back to the Fuzzy-Logic (Asus) and Easy-Tune (Gigabyte) moniker's. Abit took this Windows based BIOS tweaker even further, allowing control over the following voltages in addition to the FSB access; VGA, Vcore, and DDR. At this time µGuru is only Windows compatible so Linux users are out of luck. Personally I make such critical adjustments through the BIOS. I've noticed changes made via utilities similar to µGuru; often go undetected by most benchmarking software. Sandra SiSoftware, WCPUID, and others are simply unable to detect the changes. For the entry-level Overclocker, however; this isn't the end of the world. All one need do is use CPU-Z, to verify they're changes took. But I digress. What I found most interesting about AI7's µGuru, are its monitoring functions. Specifically a temp monitor labelled PWM. Pulling the reflective label off the chip we find a Winbond W83L950D monitor/controller chip;





The Winbond W83L95OD is basically a W83L784R on steroids. This chip monitor's a total of 8-voltages, 3-fan RPM's and 3-temp's including Intel's internal thermal diode. I've been around PC's now since 1987. I delved into the Enthusiast realm in 1999, and to date I've yet to see a motherboard which installed a chip able to monitor the power circuitry feeding the CPU. I've absolutely no criticism for this savvy engineering/marketing decision, and I admire the effort as much as I do seeing motherboard makers place heat-sinks on mosfetts. What occurred here is that Abit unknowingly revealed what I consider to be an unforgivable oversight. Once I dropped the Prescott into the Abit, I immediately noticed PWM temps soared whenever 3D or clock-cycle intensive programs ran. At just 200FSB and default Vcore, running Set@Home produced some sup rising results, as evidenced in the screenshot below;





Notice in the right lower hand corner of the µGuru EQ screen, PWM indicates 43C. While this may not seem excessive, it's relative to the SYS temp reading of just 12C. Next I raised the FSB to just 235FSB (1:1) pushing the 3.0E to 3525MHz. At no time did I raise Vcore at these speeds;





The PWM temps rise now to 72C after only a few minutes of running Seti@Home. In the first screenshot at default speed 43C may have been acceptable had the SYS (or ambient case temp) been lower (around 19C). Since the two screenshot's above were taken just 10-minutes apart at most, this is quite alarming. What we can extrapolate from data, is the power circuitry (PWM) becomes so hot, it actually influences the thermistor devoted to SYS temp. I found this happened again, and again. Whenever the PWM had risen excessively as a result of running LOAD even marginally overclocked, the SYS thermistor would also rise. Obviously, if the Case side-panel is removed (which it was), and we begin the benchmark where SYS = 12C, and several minutes later SYS = 20C, you can be certain the ambient temp didn't rise 8C or 15F in such a short period of time. At 72C/161F without any Vcore increase, it's obvious the power-circuitry on the Abit AI7 was not intended for Prescott.

And this is not an isolated incident. If you recall [H]ardOCP's article Overclocking the Prescott 2.8E Kyle had made the following observation/statement: "All of the motherboards we have used with Prescott's run hotter than we are used to. It is not uncommon to see components on the motherboard at 250F/120C degrees."

Given the architectural dialogue between chipmaker and motherboard maker, one must presume both manufacturers had an idea such problems could develop. Why then didn't they make the appropriate modifications to their Socket-478 "Prescott Ready" platforms? I believe the answer is simple. Knowing the release of a socket LGA775 Prescott was just around the corner, manufacturers chose to eschew any circuit changes on their Socket-478 platforms, even though they are purportedly Prescott compatible. The very small number of Enthusiast's who might push Prescott to such "extremes" simply didn't justify the cost of replacement parts, and redesigning of the circuits in question.

Problem is, the excessive temps seen above were not the result of overclocking in the first screenshot, but simply running Seti@Home. And 3D intensive applications stressed the board so much rebooting was common, anytime ambient temps weren't ideal. As the screenshot's above empirically verify, these problems are bound to persist, and what's more troubling is the lack of sensors on motherboards able to monitor such temps. The Abit AI7 is the only Socket-478 90nm compatible motherboard I know of, which specifically monitor's the PWM circuitry. Had it not been for uGuru's being equipped with this unique temp sensor, I'm sure this would have alluded me.


Now to determine the precise cause for these PWM temperature extremes, in the follow-up to [H]ardOCP's Prescott 2.8E piece, Prescott Tweaks Dangerous, Kyle reprints the following statements from MSI concerning the elimination of Vcore adjustment from it's 865PE Neo2 motherboard BIOS, and why;





How many Enthusiasts know about Dynamic VID? I've found some very useful info at Iamnotageek.com in an article by Martin entitled Prescott Info. In reading further through Intel's White Paper titled; VRD (Voltage Regulator Down) the 478-Socket platform is designed to regulate voltage in order to keep the Processor cool during clock-cycle intensive periods. Section 2.4 of the VRD paper states;





These features while excellent safeguards for average use, certainly present problems for Overclocker's raising Vcore. But what about circumstances where the Processor may be pushed beyond average workloads, but not necessarily overclocked such as my first screenshot exemplifying PWM temps at default speed/Vcore? Whenever Prescott is under heavy workloads, this exacerbates Socket-478 Thermal Clock Throttling "like" features. However due to the Processor's high current requirements, which are exponentially aggravated by the failure of motherboard makers to beef up the PWM circuitry, there seems to be a feedback which is becomes detrimental to the motherboard's power circuitry. The following quote was taken from the conclusion notes in section 2.8.1 from the VRD paper;

There seem to be a plethora of circuitous ad hoc thermal safety features inherent to Socket-478 power design, all conspiring to lower temperature through the regulation of voltage. Reading through Intel's White Papers pertaining to Socket-478 90nm features, it's almost redundant, how often thermal/voltage regulation is reiterated. Intel is adamant in communicating to boardmakers/end-users that Thermal Clock Throttling always be enabled;





As Martin stated in his article pertaining to line-3; "...If it's so well designed then why does it need to do crap like this?" I must confess he does make a good point. Of course no decent Reviewer would be worth his/her weight in Silver Solder if they failed to look at both sides of the argument. While these safety features might seem encumbering to the experienced Enthusiast, whom has gone to great length, and expense to install optimal cooling, what about average desktop users? Water-cooling, phase-change, and high-end after-market HSF combo's are the exception to the rule. Intel must design for worst case scenario, or minimum requirements and they have done their homework. Designing redundant thermal safety features is good engineering sense. In fact Intel knows where to throw its resources, and their commitment to the commercial market is spot-on. Business requires dependable, reliable, self-regulating Server's, and Workstations, not overclocking.


Of course this article was written for the Enthusiast, and where such features are concerned, we must go to great lengths to circumvent their effectiveness. Just about every safety feature designed into a processor, is but an obstacle to the Overclocker. So let the Overclocker be forewarned, raising Vcore on Prescott is not advisable at present, nor is voltage modification. Because of Dynamic VID, any such changes may disable the feature, or even worse see the higher Vcore as the base from which Vcore shall fluctuate. If you own a Prescott you've probably seen the range of fluctuation in its operational Vcore.

This is further complicated due to the anaemic power circuitry on Socket-478 boards claiming to be Prescott ready. I would suggest setting Vcore to Auto, instead of manually choosing the lowest voltage. On my Abit AI7 Vcore with the Prescott in socket reads 1.3875V, yet on the Asus P4C800E-dlx Vcore reads 1.365V. It seems each board's BIOS interprets this value perhaps based on feedback from its own circuitry. If you have optimal cooling, the feature can be disabled. Martin suggests opening Wpcredit, and looking for ACPI P_BLK if it there then it seems were on our way. If you have any suggestions on disabling Dynamic VID via Wpcredit, please contact Martin at IamNotaGeek.com. You may write me as well, and I'll send it along.


We can never expect Intel or AMD to make life easy for the Overclocker/Enthusiast looking overclock their processors to speeds they haven't even announced yet. Our efforts, however; bode well for them; as such performance exemplifies the processor's performance potential. Albeit just a fraction of total sales, we may be the most loyal; we are also the most fickle. Our demands and our criticisms are quite often unforgiving. Perhaps there will come a time when the growing Enthusiast market segment will warrant chipmaker's to design a processor, specifically for us. Then again where would that end? The whole point of overclocking is to exploit the manufacturing process. What fun would it be a; "...cause without a Rebel?"


// Keith Suppe (Liquid3D)

Questions/Comments: forum thread