@goriath: Please see the updates to the mod in the previous post and apply them.
My modded Voodoo5 5500 AGP is REV 2500. R22 is from ball C1 to VCC 3.3V and has to be removed.
Big attention when connecting A2 of the AGP connector to GND. Use a small thin wire and use the smallest possible blob of tin to solder it, but make sure you have a strong bond. I routed it to GND pad of C554. The TYPEDET pad of R22 I routed to GND pad of R9. You might experience issues due to the blob of solder on A2 altering the card proper position inside the AGP slot. Slightly altering the angle might help for better contact.

Edit: Tested on Tyan S2466N-4M and Epox 8K3A. I don't have other AGP 4x motherboards and from my less than 200 Euro monthly salary I doubt I'll invest in more.

m14radu wrote on 02.11.13 at 16:22:49:


I undone the mod on my 2100. After many attempts to stabilize the card, one of the two R46 pads came off and this damaged the card (no boot). The card needed a little rework to work again, but after that I left it aside since I stressed it enough

ASAP I will mod another card (let's see if I have a 2500 too).

BTW, I connected A2 directly to the VIA that I call C1, since it is the closer contact to the real C1 beneath the BGA (on the REVA2 infact, A2 and C1 are linked directly - btw it's the same since you have to put all to GND). C1 is the R22 pad which Komponent refers to. Then I connected C1 to GND. R22 was removed (on the 2100) since it is put to VCC3.3 (A9/B9) (while I confirm that resistor is put to GND on the 2300, so you could leave it in place I think or replace it with a 0R at most). R46 pads were been bridged together.

The card was working at 1.5V on a EPoX Ep-K5A3+ but unstable and yes, I used a thin wire and I done very flat solderings in order to not interfere with AGP pin contacts.

P.S. Would someone suggest where I could find an AGP riser of this kind in europe? --> http://www.attro.com/images/peripherals/AGP100.jpg

m14radu wrote on 02.11.13 at 16:22:49:

Really not much to see, courtesy of the crappy phone camera, but here you go:


Did some checking and on REV 1900 C1 is also only connected to VDDQ, not GND, like on 2500, through R22.

Did nobody else try the updated modd? Only basic skills required to do it...

Ok, got my V5 working again... damn this was quite an odyssey.

If you ever have the problem of flashing your V4/5 back and you get an error when writing the eeprom, open your bios with the tdfx bios editor, save it and DON'T rename it. Just copy it on the floppy an go ahead. If you ask me there is a very nasty bug inside the 3dfx flash tool... it does not accept names renamed under windows. Somehow it messes the checksum or something else.

Ok, back to topic... modding the card right now.

Loeschzwerg wrote on 05.11.13 at 21:23:15:


Could you confirm that R22 resistor is put to GND right?

BTW I modded my 2100 again. I wired the card using wires from an ULTRA-ATA (80 conductors) ribbon flat cable. I ended with very clean work. I tested the card on the 8K5A3+ with Windows98. I suggest you Win98 since it's perfect for debugging; if there is something wrong that you should be aware of, under Win98 it will pop up. WindowsXP is too much tolerant.

The first thing to set up is AGP 1X in the bios menu, otherwise OS will hang at loading. It seems legit since the Voodoo5 is a PCI66 video adapter so set the AGP bus to 4X is perfectly useless.

Unfortunately I can confirm the card doesn't work under stable environment. If you call the driver, the card will hang as soon as the driver will call back the hardware. For example, this behaviour it's going to happen when you change video settings in the windows display property panel, or, in game (Quake3 Arena). The card hangs randomly also under 3D environment. To lift up the pin 14 of the QUAD AND gate IC and uninstall the part of the driver chipset affecting the AGP controller didn't help so much.

I started to think the card was already unstable at the beginning, so I quickly undone the modification to try it again at 3.3V but it runs fine, so I can say in normal conditions it works properly.

Yes, R22 leads to ground. I will post a picture of the mod this afternoon.

AGP is at 1x, Fastwrites/Sideband OFF, AGP 3.0 Load Line Calibration OFF...
But these settings have no effect anyway, the card always behaves the same.

It runs the first 3DMark99 test and freezes either on the second or third.

OS is Win98 and I can confirm the card worked unmodded.

Pictures:


Note: A2 to GND is not needed on AGP4x/8x slots.

1) card also freezes running at 143MHz

2) unsoldered the power mosfet (Q5) and all the parts that come with it, no effect on stability

3) tried an AGP voltage of 1.6v, no effect