DFI X48 LT T2R: Floats like a Butterfly…

GTL - Not again!

CPU GTL reference voltage values are employed as threshold voltages by internal, on-die comparators used to determine whether each switch of an incoming data signal on either the FSB data or address busses is meant to represent a logic high or logic low (1 or 0). Adjusting these reference voltages in effect shifts the level in which the signal is seen as either high or low by the processor. As FSB is increased, the switching frequency is increased (by definition) meaning that each low to high or high to low voltage transition must occur at an accelerated rate.

This reduction in signaling margin, coupled with a more exaggerated ground bounce and overshoot, often requires manual adjustments in these thresholds if the system is to remain stable at the higher FSB speeds. Reducing or increasing the GTL reference voltages around a given percentage margin of the applied VTT by only a few millivolts can drastically affect processor stability. Imagine the processor detecting a 1 on a data transfer line when an incoming signal is in fact supposed to be low (0) - not surprising, the result is a BSOD, a complete system hang, or possibly even an undetected corruption of critical system data.

The added load capacitance of dual die processors (current Intel quad-core processors) usually requires more signaling drive as the FSB frequency is increased resulting in a need to raise VTT voltage. Any adjustment to VTT may also in turn require changes to GTL reference voltages in order to support proper data transfer margins. For 65nm processors the ratio of any GTL voltage to VTT is around 67% for both CPU and Northbridge (NB) GTL voltages; 45nm processors require that this percentage be closer to 61%-63%, again for both CPU and the NB. For example, for a VTT of 1.10V a 45nm processor would require a GTL reference voltage in the region of about 0.671V (61% of VTT) to 0.693V (63% of VTT).

DFI's approach to providing the user with ultra fine control of GTL reference voltages is unsurpassed in the motherboard industry, although working with an ultra fine scale such as this requires a significant dedication of time in order to get things right. In fact, some users enjoy this type of tweaking while others may feel it is akin to pulling teeth. In order to address this situation DFI has gone to the trouble of releasing a scale showing equivalent VTT/GTL percentages and associated BIOS settings.

We have found the scales themselves to be quite broad in range so we are providing a condensed version, taken from measurements using our own board here in the lab. The major issue here is that no two boards/CPUs are likely to behave in the exact same manner. This being the case, the small chart below is only intended to provide a starting point for any adjustments that may need to be made to your particular setup.

Easy 500FSB E8500 Setup guide (easy starting point)

In order to make things as easy and presentable as possible we decided to provide users with a video setup guide showing how to run 500FSB with an Intel E8500 processor. We cannot guarantee the settings shown in the video will work on all systems as component choices will play a huge part in how far the board can clock. Secondly, GTL values vary from board to board and across processors adding further variables in to the mix. We attempted to keep the setup as simple as possible whilst enabling all key chipset registers for performance. If there is enough demand, we may venture into a quad-core overclocking guide too - although 400-420FSB is attainable with a minimum of BIOS level adjustments, so a guide is not really necessary.