Well, I spent the whole day reading all the thread and trying to understand what google translated in english (you can figure it out) ...
First of all, kudos to Löschzwerg and (if I understood well) Eric Cantona which managed to start their REVA2 into 1.5V platforms.
In the last week I tried to understand what's mysteriously laying behind the Voodoo5 5500 AGP 2.0. Would be nice to understand (if possible) how to mod a REVA in the way it can work properly into an AGP 1.5V motherboard like seems to able to do the REVA2.
Looking at napalm databook there are three elements which seem involved into the AGP 2.0 design:
[Blockierte Grafik: http://img202.imageshack.us/img202/3228/i2k1.png]
1) The well known TYPEDET, position C1 on the BGA package
(C1'#VSA1 and C1"#VSA2 seem to be linked together internally)
Description:
AGP Signaling Level Indicator: If this input is open, the AGP interface uses 3.3 volt signaling. If this input is shorted to ground, the AGP interface uses 1.5 volt signaling. On the evaluation board, this input is connected to a resistor network consisting of a 4.7 kohm resistor to 3.3V and a 100 ohm resistor to ground. This pin should be pulled up on PCI cards.
- Indeed, on the Voodoo4 C1 is linked to the A2 pin on the AGP card connector, then connected to a 100 ohm resistor to ground. On the Voodoo5 REVA C1 is NOT linked to A2 as we already know, but still connected to a load of 440 ohm measured with my multimeter (could not be so accurate, but you got the idea). On the Voodoo5 REVA2 C1 is linked to A2, then connected to a load of 440 ohm measured with my multimeter.
2) AGP_VREF_1_5V (AVR15), position P2 on the BGA package
Description:
AGP Voltage Reference: This input supplies the switching threshold for the AGP recievers for the 1.5V case. This is derived from a resistor network consisting of a 210 ohm 1% resistor to VDDQ and a 158 ohm 1% resistor to ground. There is a 0.1 uF capacitor in parallel with the resistor to ground.
3) AGP_VREF_3V (AVR30), position P3 on the BGA package
Description:
AGP Voltage Reference: This input supplies the switching threshold for the AGP recievers for the 1.5V case (actually it shouldn't be 3.3V case?). This is derived from a resistor network consisting of a 210 ohm 1% resistor to VDDQ and a 158 ohm 1% resistor to ground. A second RC network cocnsists of a 210 ohm 1% resistor in series with a 560 pF cap to VDDQ and a 158 ohm 1% resistor in series with a 560 pF cap to ground. This is total of four resistors and two caps.
These resistor networks are present on Voodoo4 and Voodoo5 (both REVA and REVA2 - I suppose REVA1 also).
Theoretically, depending on the reading back from TYPEDET the card is able to use AVR15 circuit and select 1.5V signal if A2 is shorted, AVR30 and select 3.3V if A2 is open.
I modified the picture taken by Löschzwerg (I hope He don't take it bad) when He took off the second VSA-100 from that Voodoo Zombie. It does show two important things:
[Blockierte Grafik: http://img716.imageshack.us/img716/6199/y541.jpg]
1) P2" and P3" are clearly shorted - I assume the same for P2' and P3' (although I found that P2' and P3' are not directly connected to P2" and P3")
2) At the bottom of BGAs are located the resistor networks (blue circle) - they seem to be AVR15 circuits
3) Where are located AVR30 circuits then? I didn't find yet any resistor network which resembles to AVR30 description
BTW, I modded a Voodoo5 REVA accordingly to my findings but without success until now. I have tested the card into universal AGP slot at the moment because I don't want modifying the agp card connector or motherboard slot yet. But I think I'll have to pick up a motherboard expressly for this task.
About Voodoo5 REVA2:
I have one of these card too. Mine it's dated 3500 as many others. In the last days I tried the card with two SiS645DX based motherboard, but it seems I'm bit out of luck; I'm annoyed to have found these motherboards as not working 
, maybe the AGP slots are defective or dunno .. the last time I used they were ok 
Currently I'm using an EPoX EP-8K5A3+ and as someone else previously said already the card starts but at 3.3V (while Voodoo4 keeps working at 1.5V, lol!) - I have other motherboards with universal AGP slots but they are all based on the ViA KT333, so it's useless.
Intel i875 chipsets, sadly, aren't gonna work:
About 1 year an half ago I tried my REVA2 with an ASUS P4C800-E Deluxe (intel i875P, AGP3.0 slot). Well, as soon as I try to start the system, it keeps to power on and off continuously and so quickly that seems underpowered. Because of this issue I immediately quit the experimet, but no damage has occurred either to motherboard and REVA2. Today after reading the thread I tried again, but with the same weird result. This time I tried to measure the AGP voltage during the test; readings vary very much since voltage keeps dropping repeatedly but I would say it can't go over 0.6V. Dunno if the same chipset on any other motherboards will act differently, but with this mobo is a no go.
Now... I did some research and found the following statement from Hercules:
Zitat
"Regarding Pin A2 signal, this pin signal is read only on AGP universalbus (with no AGP 4X notch), and on AGP universal bus, our board switch to AGP signals in 3.3 volts, you can read this spec on AGP 2.0 specs (chapters 4.3.5 and 4.3.6), so this Pin A2 signal is not relevant for a test on a non universal AGP bus, as on DFI VC11, and as on other i845 motherboards.
The only conclusion one can make from Pin A2 signal is that, on a AGP universal bus (without the 4X notch), you use a 3.3 volts, but that is a signal voltage only for an AGP universal bus, it is not the signal for a i845 or i850 AGP bus."
What does this tell us? If I plug the V5 AGP4x in a board using AGP universal slot it gives 3.3v.
Exaclty, the card seems to act ALWAYS like you said if plugged into an universal AGP slot.
But why? The Voodoo4 starts ALWAYS with 1.5V signal if the motherboard is capable of that...
Zitat
If you ask me the Rev A2 card is safe to be used in any kind of AGP board, beside AGP 8x only (AGR or XGR...). EDIT: 3dfx did everything right here. The Rev A2 work with 3.3v in AGP1.0 and universal boards; 1.5v on AGP4x notched boards.
My experience with the P4C800-E Deluxe seems to confirm your thought (although nothing bad happened). But an AGP 3.0 slot shouldn't be backward compatible with 2.0 specifications? Looking at AGP3.0 documentation (page 39) an Universal 1.5V AGP3.0 Motherboard has an AGP slot 3.0 backward compatible with 2.0 standard, while an AGP3.0 Motherboard has a slot working only at 0.8V singaling (4X/8X) and has additional electrical ID to prevent AGP 1.5 V operation, but the fact is that I never found a motherboard which hadn't such compatibility and manufacturers never referred to a *universal AGP 3.0 slot* but simply to an AGP 3.0 slot (8X/4X @0.8V or 4X/2X maybe 1X too @1.5V). The Abit NF7 has a *so called* AGP 3.0 slot, but is able to feed the adapter with 1.5V singaling and this motherboard has proved to work with the REVA2. But the P4C800-E Deluxe doesn't seem to be different. The manual states clearly that The AGP 3.0 slot is able to feed 1.5V cards too and the Voodoo4 works indeed...
BTW, sorry for the long post 
). But the CPU power supply line is poor, only 2-phases. Someone may observe that for Northwood based systems it's enough, while 3-phases design is needed for platform designed around Prescott core. The BIOS is designed to keep the board alive and make the basic adjustments to run the board properly, nothing less nothing more.
Just possible AGP feature set and we all know the V5 can't do anything more than PCI66.
