It's amazing how this thread has just died in the last month.

Anyway, just thought I would share some GTL tuning experiences.

The first 2 GTL settings affect the Data Strobe (ref 0/2) and the Address Strobe (ref 1/3), respectively, NOT the cores or pairs of cores themselves as I had thought. In fact, these 2 GTL settings have nothing to do with the cores directly, but everything to do with the FSB and how the FSB detects what the cores are telling it.


Here is an analogy: If we think about this relationship between the CPU, FSB, and the MCH as a movie scene in a busy downtown setting in which the actor is speaking quietly, the scene of the movie becomes the entire signal that the CPU is sending to your stereo (the FSB) in order to transmit the actor's voice (the actual signal of interest) to the viewer (the MCH...could also be another core in the processor, but will assume for simplicity's sake that this is not the case here.)

Now since the scene was pre-recorded, the ratio of the volume of the actor's voice to the background noise cannot be adjusted. We can, however, turn up the volume of the stereo (vFSB) to make what they are saying more evident, and this works well in most cases. But at one point in this scene, a car horn (background noise) honks at the same time as the actor says a word, making understanding the word impossible no matter how loud (vFSB) you turn up the stereo.

So, you go to your bass and treble adjustment knobs (GTL refs 0/2 and 1/3) and adjust the frequencies (this isn’t exactly what happens, but works well) so that the noise introduced by the car horn is reproduced less by your stereo in relationship to the actor's voice and now the word can be understood by the viewers (the MCH).

But one of these viewers (the MCH) is a senior citizen whose hearing is going and sometimes cannot understand the words (doesn’t understand the signal sent by the FSB and therefore cannot transmit it to the ICH, RAM, etc.). But luckily, he wears a hearing aide (an amplifier-vMCH; and a tone adjustment knob GTL MCH REF) to assist him, but the components of the hearing aide are of lesser quality than those of the stereo (FSB) and provide only coarse tuning, but are effective.


Now back to computers:
As the speed of the FSB increases, so does the amount of background noise around the signal, now we can increase the vFSB to make this signal louder, or we can tune the GTL refs to better detect the signal, OR we can do both.

At "Auto" GTL settings, my Q9550 at 6x maxed stable FSB out at 472 with vFSB and vMCH maxed out, 2 dimms of ram fairly loose and @ 1:1.

Started tweaking the first 2 GTL REFs, gained stability, increased FSB, lost stability, repeated...now I hit a wall around 490, but it was different from the points of instability that were corrected or induced by changing the first 2 GTL settings. Those experiences resulted in windows freezing up at various points during loading (the longer it takes to freeze the better and adjusting up or down from the optimum will shorten this time), but this time the computer just failed to do anything other than reset or freeze at black screen once all of the POSTs completed. So I tried my hand at the GTL MCH REF setting and now FSB stable at 500 MHz @ 6x, but can’t get it any higher.

Started bumping up the multi as well as the vCORE. Not enough vCORE to remain stable at 500x8 with current cooling (liquid cooling on the way!) and wont run long at 8.5 multi. So I backed off the fsb some for now and running “stable” at 490x8 with vCORE 1.4375 in BIOS :: 1.41 in Everest @ idle :: 1.38 in Everest under load but still gets really hot, so no super long stress testing yet.
Will update next week once liquid CPU and NB cooling are installed.