The reason I wrote this small article... is just my habit (quite a useful habit, actually) to run the RMClock utility in the background while I run tests on Intel Pentium 4 platforms — just to make sure that a processor does not enter the throttling mode and thus the results obtained are reliable. And that's what I found out — after a couple of minutes of running tests on Intel Pentium 4 670 (3.8 GHz) with ASUS P5WD2 Premium (I had just flashed the latest BIOS Version 0422 dated 29 July), the processor started throttling and dropped its performance to... 105.6%!
It goes without saying that we couldn't ignore such an unexpected behavior and decided to investigate. At first we were surprised even deeper — we just restored the RMClock window and had a look at its main tabbed page (in order to avoid noticeable oscillations of "throttling clock", when a processor is not loaded, we temporarily disabled C1E).
The picture is truly impressive — the "throttling clock" of the processor turns out much higher than its reference clock (of course, the throttling efficiency cannot possibly exceed 100% — the CPU core just operates at a higher clock than the part of the processor with TSC). We found out the reason for this phenomenon rather quickly — it turns out that the motherboard BIOS puts Startup FID one point too low. In our case it's 18x instead of 19x. The same thing happens when we set other multipliers in BIOS — from 18x to 15x (14x remains for ever one, as it's the absolute minimum for Prescott cores).
So, what is actually going on? The CPU clock is supposedly 3.8 GHz — that's what the operating system and various utilities (CPU-Z, WCPUID, etc) will tell you. But this is just some "reference" CPU clock — the clock of TSC, which lost much of its relevance as a core clock indicator with the release of Prescott cores. The only way to obtain it is to... set the FSB clock to 3800/18 = 211 MHz. But the real core clock is much higher — 211x19 = 4010 MHz. Whether it's good or bad is not up to us to decide. But what's really bad is that it happens, when we set the FSB clock in BIOS strictly to 200 MHz (we don't use any "intellectual" overclocking functions like AI N.O.S.). That is we set one thing, but get quite a different one. We got a constantly throttling processor at best (you will hardly find a processor, which is guaranteed to run at 200 MHz higher than its standard clock and not getting overheated — also taking into account the quiet design of the motherboard). Or we got an unstable system at worst (don't forget that the increased FSB clock results in the increase of memory frequency).
All the above said refers to the operating mode with enabled Enhanced Intel SpeedStep (DBS), which is for some unknown reason disabled by default in BIOS Setup. It's not alone, Execute Disable bit, Automatic Thermal Protection, and Enhanced Halt State are also disabled... When it's disabled, the picture is a tad different.
In this case the processor is not overclocked — we have the honest 3.8 GHz, but they are not obtained in an honest way. By the way, results of RMMA tests also speak of the "dishonest" overclocking (that is unauthorized overclocking)
The 7177 MB/s maximum real memory read bandwidth exceeds the theoretical limit of 6400 MB/s for the 200 MHz FSB and can obviously be obtained only by overclocking.
And now the most important part. Just change the FSB clock in BIOS Setup — set it to 199 MHz (but not to 201 MHz or higher!), and everything falls into place.
In this case the motherboard BIOS plays no tricks with the start-up value of the CPU multiplier — it is set correctly to 19x, so we get the correct FSB clock. But even in this case the facts of life spoil the seemingly nice impression. Just several minutes of tests are enough to make sure that the processor runs on the point of throttling (RMClock is constantly reporting about throttling on the 100% level, which means that the temperature threshold of the CPU overheating protection is exceeded). That happens with the standard Intel cooler in an open PC case...
Mind it that all the above said refers only to the new BIOS Version 0422. We didn't witness such behavior (I mean the FSB clock) with the previous BIOS 0205. It's hard to say whether it's a bug or a feature, of course. But we are inclined to consider it an original feature that serves to establish performance superiority of this motherboard among the similar models. But this is achieved by dishonest means and, besides, it may lead to undesirable consequences described above. If it were a bug (that is the manufacturer suddenly unlearnt to specify the correct CPU multiplier), its mysterious disappearance in case of a tad slower FSB clock would be absolutely impossible to understand.
