gigabyte MB - what the hell is 'cpu reference voltage'

G

graysky

Gawd
Joined
May 6, 2007
Messages
620
Guess I'm used to DFI naming for various voltage settings because I have a GA-G41M-ES2L and have no idea what "CPU Reference Voltage" is... what is this?

Code: 
CPU Vcore
CPU Termination
CPU Reference

CPU Vcore is obviously the core voltage for the processor.
CPU Termination I am assuming is the VTT.
CPU Reference = ?

It has the following settings:

Code: 
Auto
Normal
0.738V
0.756V
0.780V
0.805V
 
Last edited:
HenryBravo said:
Here is a good explanation.
Click to expand...

Actually that is not helpful at all. However a very nicely executed fail compared to most.

-----------------

OK I am going to give it a shot. Somewhat simplified. And also somewhat of a guess as you have already figured out, there is not a clear public explanation from anyone of authority that I can find. The best info I have found is the Intel data sheets and this: http://www.thetechrepository.com/showthread.php?t=87

Dear Gigabyte, for the measly cost of a new X48 board I will rewrite your crappy manual's bios section and provide it to you with all rights, so that your customers think better of you and your products. (not holding my breath)

What we are doing here and a key thing to remember is we are attempting to modify the voltages on the CPU to MCH(aka Northbridge) data buss to compensate for increased CPU frequencies (overclocking) outside of Intel's stock specifications. Changing the buss frequency (overclocking the CPU via FSB clock increases) causes additional signal reflection and signal/clock skew (aka ringback) due to the capacitive nature of the devices (I/O transceivers/receivers buffers on CPU and MCH) on the buss.

Here is the key from the C2D Intel data-sheets.
----------
The GTL+ inputs require a reference voltage (GTLREF) which is used by the receivers to
determine if a signal is a logical 0 or a logical 1. GTLREF must be generated on the
motherboard (see Table 15 for GTLREF specifications). Termination resistors (RTT) for
GTL+ signals are provided on the processor silicon and are terminated to VTT. Intel
chipsets will also provide on-die termination, thus eliminating the need to terminate the
bus on the motherboard for most GTL+ signals.
----------

A valid logic 1 or O on the FSB buss is determined if the voltage of a data bit pulse is above or below a certain voltage level (GTLref) with a +/- 10% "dead band". This GTLref voltage is derived from Vtt (CPU termination voltage) by a voltage divider circuit. So we have two voltages, Vtt is the starting voltage if you will, and GTLref is the derived voltage which determine if a bit is a one or a zero.
GTLref is stock set at 2/3 of Vtt

I believe GTLref is what Gigabyte is calling CPU reference voltage but I have no way to confirm.

Stock Vtt is 1.2V
Stock GTLref is 2/3 of Vtt = .804 <<<---- Notice how your settings options are right about this amount or slightly lower.
talking now about the +/- 10% "dead band" 10% of .804 = .08 so ...
A voltage greater than .884v on the buss (.804 + .08) is a logic high or a "1"
A voltage less than .716v on the buss (.804 - .08) is a logic low or a "0"
Anything in between .884 and .716 is invalid and an error with very bad results.

Again, normally, GTL is derived by a resistor network and is always 2/3 of VTT. Or said in Gigabytes way, normally CPU Reference Voltage is 2/3 of CPU Termination Voltage.

OK the idea is to be able to move these voltages around to help compensate for increasing the FSB frequency.

If you adjust Vtt you also adjust Vgtl (normally Vgtl is 2/3 of Vtt).

If you raise Vtt much in an attempt to compensate for a poor power supply or motherboard component variations or if you just want to give it some more oomph (drive strength) (1.55 is absolute max per Intel data sheet) Vgtl rapidly goes out of spec and exceeds the .884V. Doing the math a Vtt of about 1.326 will make signals that would have fallen in the "dead band" a valid logical one. Likely not a good thing.

Notice how your options are mainly to reduce CPU Reference Voltage below the the .804V level. But are "in the ballpark". This is what leads me to beleive this is the "lock down" for GTL reference voltage that Gigabyte is calling the CPU reference voltage.

I think the idea is that you can increase Vtt to ensure it does not "droop" due to the demand of high current under load and when OCIng (give it more oomph), but set/lock the GTL voltage (CPU reference voltage) below or near spec so the I/O transceviers on the buss do not freak out *. (CPU load should not normally affect Vtt as that is the reason we have Vtt and Vcore as seperate voltages but who knows how well the two are truly independent and the power supply powers both and the board may or may not regulate Vtt as well as one could hope even if Vtt has its own on board regulation circuit independent of the main CPU voltage regulation circuitry. ) Variations in components, temp drift and a zillion other things are all variables making each and every board and computer system individually different.

So I recommend (for starters) to raise Vtt a bit for more oomph but set the CPU reference voltage to .805 (clse to stock) and see where that gets you. After that you are on your own as every chip and motherboard (even with 1% resistors in the GTL voltage divider circuit) are different. Temp drift and God only knows what else will make setting these to the best for your individual board a challenge. A really good O-scope on the data buss would be the only way other than trial and error to know what the hell is actually going on.

Before I get flamed, the clever Intel people even go so far as to invert 1 and 0's on the buss depending on the amount of 1's so that normally most data is transmitted as a lower voltage "0" to reduce power comsumption and device heating. As long as the transceiverrs on both ends know the data has been inverted it causes no "confusion".

Disclaimer:
The above is an very expensive well educated guess, thats all. I could be completely wrong, it happens often enough.
* BTW "freak out" is a valid engineering term for the results seen in any system by the propagation of invalid data.
 
Last edited:
BillParrish said:
However a very nicely executed fail compared to most.
Click to expand...
Thanks, I thought it was nicely executed as well.

My point (in case you missed it) was that a little bit of research goes a long way. I bet when you were first learning about CPU reference voltage, you hit up a search engine or two.

HB
 
Cool, thanks for the info. The list seems to adjust automatically for the VTT you select giving you four options over the range of VTTs you chose: 0.62x, 0.63x, 0.65x, and 0.67x.
 
I bet when you were first learning about CPU reference voltage, you hit up a search engine or two.
Click to expand...

You would loose that bet. Intel.com is sufficient and I know how to find it all by myself.
 
You must log in or register to reply here.
Back
Top