- Joined
- Apr 15, 2001
- Location
- Turku, Finland
X-bit labs has an article about GF3 voltage tweaking that is often used as a reference for these mods. Although it has valuable information about making the mod, one should exercise caution when applying the information in the article. The resistor values or voltages mentioned in the article are not necessarily correct ones for other cards but just those two used.
I have an Asus GF3 V8200d which is overclockable to 250/537MHz. Because GF3 is well balanced I was feeling that I could use some more voltage for memory to get its speed to same level as GPU's. After looking at the resistors on the card, data sheet of voltage regulators and making some measurements with multimeter I had to note that resistors Asus had used were quite different from those mentioned in the article.
The reason why the core in my card can go so high is obvious: it has core voltage of 1.65V - not the 1.48V mentioned. If I was to make a core voltage mod for the card, use of 820 ohms resistor probably would heat up the card more than I would like because of high core voltage.
Memory regulator resistors on my V8200 are quite different from resistors on the card used in the article. Measured voltages are anyway in the vicinity: 2.61V and 3.46V. Using 820 ohms would give me voltages of about 2.80V and 3.78V. Giving 2.80V for buffers on the memory of my card doesn't do any good - the memory performance is limited by the circuits using 3.46V. Actually I will have severe artefacting in 3D Mark2001 if I raise 2.5V voltage high (although memory temperature is only 28°C).
Because I couldn't resist the temptation of doing the mod I did it by using 200 ohms resistance and 5k trimmer in series for 2.5V and 500 ohms resistance and 5k trimmer in series for 3.3V. I also used jumpers so that I can switch either of the mods off. I soldered the circuitry on a small PCB which I attached with a screw to the display card.
I then measured dialed resistances while jumpers were off, closed the circuit with jumpers, repeated the measurement for new divider R (unreliable because of other components on the display card) and booted up the machine to BIOS. Then I measured the voltages between ground and +-legs of corresponding output capacitors. Turned machine off and dialed new values. This gave me following tables and information about how much I should turn the trimmer for wanted effect.
2,5V
added...new......new
R/ohm...R/ohm...U/V
0.........98,5......2,61
4160....97,5......2,65
2500....95,6......2,67
1880....94,5......2,69
1740....94,0......2,70
..730....86,6......2,83
..488....82,8......2,93
3,3V
added...new......new
R/ohm...R/ohm...U/V
0.........124.......3,46
5170....121,9....3,52
3250....120,0....3,55
2450....118,6....3,58
2360....118,5....3,58
..830..... - .......3,78
..745....107,0....3,83
Overclocking testing with 3D Mark 2001 was not so encouraging; I had to raise 3.3V voltage to about 3.9V for stability at 560 MHz and to get rid of visual defects. As mentioned raising 2.5V voltage didn't do much. This 23 MHz o/c from 537 MHz gained only 37 3D Marks when rest of the computer was not o/ced. I didn't continue further.
So the conclusion would be that it didn't gain substantial benefit in overclockability but was fun to do. My GF3 is now more prone to break because of the soldered wires on it - or burned if I will continue the testing with high voltages. This is not a worthwhile mod in my opinion and risky if one doesn't know what he is doing.
I have an Asus GF3 V8200d which is overclockable to 250/537MHz. Because GF3 is well balanced I was feeling that I could use some more voltage for memory to get its speed to same level as GPU's. After looking at the resistors on the card, data sheet of voltage regulators and making some measurements with multimeter I had to note that resistors Asus had used were quite different from those mentioned in the article.
The reason why the core in my card can go so high is obvious: it has core voltage of 1.65V - not the 1.48V mentioned. If I was to make a core voltage mod for the card, use of 820 ohms resistor probably would heat up the card more than I would like because of high core voltage.
Memory regulator resistors on my V8200 are quite different from resistors on the card used in the article. Measured voltages are anyway in the vicinity: 2.61V and 3.46V. Using 820 ohms would give me voltages of about 2.80V and 3.78V. Giving 2.80V for buffers on the memory of my card doesn't do any good - the memory performance is limited by the circuits using 3.46V. Actually I will have severe artefacting in 3D Mark2001 if I raise 2.5V voltage high (although memory temperature is only 28°C).
Because I couldn't resist the temptation of doing the mod I did it by using 200 ohms resistance and 5k trimmer in series for 2.5V and 500 ohms resistance and 5k trimmer in series for 3.3V. I also used jumpers so that I can switch either of the mods off. I soldered the circuitry on a small PCB which I attached with a screw to the display card.
I then measured dialed resistances while jumpers were off, closed the circuit with jumpers, repeated the measurement for new divider R (unreliable because of other components on the display card) and booted up the machine to BIOS. Then I measured the voltages between ground and +-legs of corresponding output capacitors. Turned machine off and dialed new values. This gave me following tables and information about how much I should turn the trimmer for wanted effect.
2,5V
added...new......new
R/ohm...R/ohm...U/V
0.........98,5......2,61
4160....97,5......2,65
2500....95,6......2,67
1880....94,5......2,69
1740....94,0......2,70
..730....86,6......2,83
..488....82,8......2,93
3,3V
added...new......new
R/ohm...R/ohm...U/V
0.........124.......3,46
5170....121,9....3,52
3250....120,0....3,55
2450....118,6....3,58
2360....118,5....3,58
..830..... - .......3,78
..745....107,0....3,83
Overclocking testing with 3D Mark 2001 was not so encouraging; I had to raise 3.3V voltage to about 3.9V for stability at 560 MHz and to get rid of visual defects. As mentioned raising 2.5V voltage didn't do much. This 23 MHz o/c from 537 MHz gained only 37 3D Marks when rest of the computer was not o/ced. I didn't continue further.
So the conclusion would be that it didn't gain substantial benefit in overclockability but was fun to do. My GF3 is now more prone to break because of the soldered wires on it - or burned if I will continue the testing with high voltages. This is not a worthwhile mod in my opinion and risky if one doesn't know what he is doing.
