ATI 4850/4870 voltmod thread

[@lex]

As long as all new posts have vanished, I'll ask once again: is there pencil GPU Vmod for this card?
http://ceemic.pri.ee/hardware/hd4850_2/
I'm sorry for asking so many times, but I really want to make that mod.
Thanks in advance.

[ceemic]

I posted this allready, but database loss again...
http://ceemic.pri.ee/hardware/hd4850...50_mod_pen.JPG

[@lex]

Thanks a lot!
Do you know, what resistance corresponds to certain voltage?

[Scrizz]

can someone help with this guy
xfx 4850? THX

shot of the back


front

[ceemic]

Scrizz, do a close-up macro from that area:

And say the markings on that chip.

PS. Please use thumbnails or links instead of these fuge pics on Your post!

[@lex]

I posted this allready, but database loss again...
http://ceemic.pri.ee/hardware/hd4850...50_mod_pen.JPG

Higher resistance means lower voltage? But any definite resistance will give voltage higher than stock? I'm asking because I don't want to fry my card.

[ceemic]

Lower resistance means higher Vcore, yes. But the correlation ins't mathematically linear probably.
Lower the resistance by 10% and see how big Vcore jump was.

[largon]

Resistance-voltage correlation should go about like this:

 vcore × [Ω orig]
──────────────────  = vcore new
     [Ω new]

[@lex]

Resistance-voltage correlation should go about like this:

 vcore × [Ω orig]
──────────────────  = vcore new
     [Ω new]

That would be too simple
From my experience such things are never linear.

[@lex]

http://ceemic.pri.ee/hardware/hd4850...50_mod_pen.JPG

Left marked pad on the picture is GND?

[ceemic]

Yes: http://ceemic.pri.ee/hardware/hd4850_2/hd4850_02g.JPG

[@lex]

Oh, sorry, I should have watched ALL pictures before asking.
Anyway, it's clear to me now. I'm waiting for VRM heatspreader, after I'll have it I make VM and post results.
Thanks once again.

[aGeoM]

Scrizz, do a close-up macro from that area:

And say the markings on that chip.

PS. Please use thumbnails or links instead of these fuge pics on Your post!

Hi ceemic

I'm also interested on this voltmod for the XFX 4850 XXX, more on pencil mod than hardmod, but any of the 2 it's ok.

Here is the close-up photo.



I hope it's ok. The marking of the chip are visible.

Thanks.

Paulo

[ceemic]

aGeoM, XFX HD4830 uses the same IC and has pretty much the same layout. Check that thread: http://www.xtremesystems.org/forums/...d.php?t=222852

[aGeoM]

Thanks, I will follow that thread.

[Scrizz]

Thx ceemic and aGeoM
I don't have a camera so...
ThX!

[@lex]

I have an offtopic question here. Help me, please.
How do you people think which RAM heatsink is better: from AC S1, or these copper Thermaltake?

Each has obvious pros and contras... TT is very heavy, it can unstick; it's fins are thick, space betwwen them is tight - hard to create good airflow. But it's copper. AC heatsink has a little bigger fins, they are thin; but it's Al... I don't know which to choose

[celemine1Gig]

The correlation between resistance change and vcore should indeed be linear, as there are only resistances in the feedback loop, AFAIK, and resistances are linear parts.
The only thing non-linear is the resistance/turn curve of a potentiometer, which gets highly non-linear in the lower reistance regions.

[@lex]

The correlation between resistance change and vcore should indeed be linear, as there are only resistances in the feedback loop, AFAIK, and resistances are linear parts.
The only thing non-linear is the resistance/turn curve of a potentiometer, which gets highly non-linear in the lower reistance regions.

I've read some basic things about PWM DC-DC voltage convertors. According to simplified scheme, that resistor should be the part of voltage divider, which normalizes output voltage to compare it to internal reference voltage. In that case correlation resistance/output voltage should really be linear.

[makaka]

HI there ;
can any one help with this card : palit sonic non reference design ?
if possible the pencil mod for the gpu , if not just the GPU soldering mod .

[celemine1Gig]

Have a look at these two pics:

http://www.hs-augsburg.de/~ingmar_k/vgpu.jpeg

http://www.hs-augsburg.de/~ingmar_k/vmem.jpeg

And then give me the details.

[makaka]

Thanks
Hi there i tried to write the marking but this was pretty hard since the latter are almost erased , for the VGPU image it was impossible to read the hole marking i just wrote what was possible , sorry .
The question is is pencil vGPU mod possible ?.
EDIT : after more checking the gpu chip marking in 90% : ncp 5388

[celemine1Gig]

A trick to make the markings visible is taking a q-tip, rubbing some white thermal paste on it and then smear that onto the chip markings. Then remove the thermal paste again, maybe using another new qtip. Now the white thermal paste should've left some residues on the chip, which should make the markings a lot clearer.

And concerning the pencil mod question:
I will tell you what resistances are needed and between which pins. I won't tell you exactly where to pencil. For one, because pencils should be used for writing and not for vmodding hardware. Anyone who thinks that a pencil is perfect for influencing a feedback line and getting long-term stable results that way, should seriously rethink that.

So, please check the markings again to make sure they are correct.

[makaka]

Hi there ;
thanks for your reply
I confirm the chip is ncp 5388
note : i tried the volt mod ((just like the sapphire card using the same voltage controller)) to the pin 19 (50 h vr) but this result in black screen , i tested the vgpu and found it was 0 . So the point was wrong . could u please check with me .
here the point market in blue.

[celemine1Gig]

You know that you probably killed your card?
That's why you always need to double/triple check. Just because a circuit looks like another one, or uses one identical chip doesn't mean that it is the same. Will have a thorough look at it when I find the time.

Edit: Measure resistance between pin#19 (Vfb) and pin#14 (AGND), with the system completely disconneced from the wall. Of course you can use alternate soldering points, that are direcly connected to these pins, but make sure that they really are.