ATI 4850/4870 voltmod thread

[[XC] gomeler]

Odd, my 8800GT with Samsung GDDR3 modules responds rather nicely, always thought it was the Qimonda modules crapping out. I'd also like to know if the RV770 power circuit has the same 1.7v OVP and if there is a sensitive OCP circuit in place? Anyone know if RV770 coldbugs badly also? All questions I could answer in a week but I'm impatient.

[Clemmaster]

Could someone measure the resistor between the pad near C601 (Vgpu) and the resistor (capacitor?) that is just beside.
I used to read 44k but once I put soldering on the pad I have now 0 ... I don't want to put the card to test I'd rather have a double check for both 44k or 0ohm. I don't see any soldering between them so I'm a bit confused.

Thank you

[jason4207]

It was the memory controller on the 8800GT/GTS w/ Qimonda that didn't like additional voltage. It wasn't the memory itself. The memory and memory controller shared the same voltage line...that was the problem.

I'm not sure if the 4850 is set up the same way, though. Hopefully, it'll respond nicely to some extra juice!

Odd, my 8800GT with Samsung GDDR3 modules responds rather nicely, always thought it was the Qimonda modules crapping out. I'd also like to know if the RV770 power circuit has the same 1.7v OVP and if there is a sensitive OCP circuit in place? Anyone know if RV770 coldbugs badly also? All questions I could answer in a week but I'm impatient.

You are correct about the Samsung chips. My GTS w/ Samsung loves voltage. It was theorized that it wasn't the Qimonda chips themselves, but the memory controller that was causing issues on the cards that had Qimonda memory. The reasoning being that turning down the voltage actually helped one obtain higher clocks, and at the brink of stability the timings on the Qimonda chips could still be tightened w/o having any adverse effect on stability. It seemed as though something else must be at fault. The memory controller was spec'd at 1.8v, but going from 1.2ns to 1.0ns Qimonda RAM only involved pumping up the voltage from 1.8v to 2.0v on the RAM (the controller had to come along for the ride since they shared the same rail). The Qimonda RAM did fine w/ the extra voltage, but the controller wasn't stable anymore. Iirc the spec sheet on the Qimonda has them rated up to 2.5v!

Hopefully, the Qimonda RAM on the 4850 will respond well to extra voltage.

[Boogerlad]

Server 2003 R2



Nah, this is figure I entered for the modded bios before flashing.

Download Link: http://myfreefilehosting.com/f/c6758073d2_0.06MB

then isn't the extra 5 gigs of ram wasted cause your on 32 bit os?

[gurusan]

So, with all that said, is there a pencil vdimm mod on the 4850

[JamesBong420]

hoping to break 950mhz tonight!! and some 3d01

to bad i suck at os tweaks..!

[Zucker2k]

then isn't the extra 5 gigs of ram wasted cause your on 32 bit os?

I'm quad booting:

Windows Server 2003 - supports 32GB (with PAE switch of course). I did run the test without it as my x-fi card throws fits with more than 4GB of ram.

Windows XP SP3

Windows XP x64 SP2

Windows Vista x64 SP1

This is just the way I like it.

[Clemmaster]

There's something strange with my Vmod : When I measure resistivity between the pad near the C601 (Vgpu point) and the second pad in the pencil mod, I get the resistivity of my VR :??:

So to improve the Vgpu I need to set the VR to less than 5.2k (4.7k for 0.11v increment). So why is there 100k mentionned? A 10k would have been more accurate don't you think?

[JamesBong420]

http://www.aoaforums.com/forum/hardw...mod-guide.html
read the 2nd part below the "OR" that should help..

[Boogerlad]

the vgddr mod is kinda confusing. do i solder either whitish point to ground? also, where's the measuring point for vgddr?

[darkskypoet]

I have just done a pencil mod and it is confirmed working....it's 1.16 idle now and 1.24 load. will post a pic in a second

Just to confirm:

Just the pencil mod will work, as in no need to solder the res as specified on the photo?

Also, instead of a pencil, as the graphite will wear as time goes on; can i use a trace pen like I've done on all my old school cpu pencil mods?

Thanks

[gurusan]

yes, I just reused demoneyez pic, that's why it says to solder...just ignore that part. And I don't see any problem with using a trace pen as long as you monitor the resistance very carefully.

[darkskypoet]

I see.. the traces from the pen could provide less resistance between the points then the graphite from the pencil? If so would this serve to increase the effective gain in voltage?
(I don't have a MM so i may just stick with pencil then, although thickness of the layer would do the same... hmmmm)

Idea though, is to connect those two points as shown in blue?

[rym]

i am having trouble flashing the bios on my 4850 's
i flashed with ati flash . i dloaded a bios from one of these threads . rv770. 1.26v and 725/1125 . i changed to 725 1150 1.26v . i have to flash atiflash -f -p 01 7251150.rom and it worked . did the same using 02 and it seemed to flash.. both cards are in cf at same time of flashing. now in vista 32, gpuz 2.4 show clocks at stock.. what am i doing wrong.

Randy

[gurusan]

I see.. the traces from the pen could provide less resistance between the points then the graphite from the pencil? If so would this serve to increase the effective gain in voltage?
(I don't have a MM so i may just stick with pencil then, although thickness of the layer would do the same... hmmmm)

Idea though, is to connect those two points as shown in blue?

i would not even think about attempting this mod without a multimeter

[jimmyz]

i am having trouble flashing the bios on my 4850 's
i flashed with ati flash . i dloaded a bios from one of these threads . rv770. 1.26v and 725/1125 . i changed to 725 1150 1.26v . i have to flash atiflash -f -p 01 7251150.rom and it worked . did the same using 02 and it seemed to flash.. both cards are in cf at same time of flashing. now in vista 32, gpuz 2.4 show clocks at stock.. what am i doing wrong.

Randy

Try to flash again, if it gives you the pop-up that the bios is already loaded then reboot and check GPU-Z again. mine sometimes shows stock clocks but going to the sensor page or switching adapters causes it to read properly.


I agree with gurusan, a multi can be bought for as cheap as 5 dollars these days, no sense trying any mod like this with out it.

[jason4207]

I see.. the traces from the pen could provide less resistance between the points then the graphite from the pencil? If so would this serve to increase the effective gain in voltage?
(I don't have a MM so i may just stick with pencil then, although thickness of the layer would do the same... hmmmm)

Idea though, is to connect those two points as shown in blue?

It's not a very smart move to do it w/o a DMM. You can get a DMM for $20. Much cheaper than replacing a $150-200 gfx card.

Be careful!

[darkskypoet]

Understood, there is a bit more to it then I thought. I needed an MM anyway

Does any one have a limited table of values for what resistances will net you x voltage? As in a bit higher, or lower? Or is there a formula for figuring this out, for these two points.

Thx

[jason4207]

Can someone tell me where the vmem measure point is?

I know you guys know. I hate to stick my probe in there while the card is alive. I'd prefer to just solder on my read point wires, and go from there.

Hook me up!

[jason4207]

the vgddr mod is kinda confusing. do i solder either whitish point to ground? also, where's the measuring point for vgddr?

After looking in the 3850/3870 thread I'm pretty sure you solder the 2 legs of the VR to the 2 points on the pic. Perhaps one of them is GND?

I'm also looking for that measure point. I think I have it narrowed down to 2 different points based on the 3850/3870 thread, but I'd like to hear from direct experience...

[rym]

ok i see one card has the clocks higher but the other wont show.. winflash says the bios is loaded already.. do i need to disable ccc or mom ?

this is the active card the other inactive shows clocks corectly. 725 1125 in gpuz

[jason4207]

OK...I'm impatient. I stuck the probes in there (man this card is hot!), and my guess was correct. The points below show up as 2.01v on the DMM, so I'm pretty sure they are the measure points for the GDDR. Someone correct my drawing if anything is incorrect, please.



Pic courtesy of Techpowerup


Also, I want to rig the fan to run 100%. Are the red and black fan lead wires the positive and GND? I'm pretty sure they should be, but I don't want to fry the fan. Yellow for sensor, and blue for PWM?

[Specialfighter]

in hardwareluxx we have tested the vmod with a 25k variable resistor, but 25k is to big its not accurate enough, i think 10k is the best choice.

is the vmem mod still working ?

[darkskypoet]

Question for those that lowered the clocks in 2D mode with RBE. I have changed Clock Info 1 to 350/500... And upped the fan speed. Fan speed works well, but neither CCC, nor GPU-Z are showing lower 2d clocks.

1) Do i need to modify both Clock Info 1, and 2? Or are there more spots that must be changed to drop 2d settings.
2) I've seen someone mention a 160 / 500 setup for 2d clocks. It sounds rather low, however is there a 'right' setting for this? As in something used in a MFg's bios? (again i seem to remember an MSI bios being mentioned with lower 2d clocks, but I've been spinning through so many forums, i am not sure If I could track it back down again)

3) With stock voltage, what sort of clocks are people getting? (Have to wait till i buy my DMM for the pencil vmod.)

(tips hat to all)
Thank you.

Edit: One thing I noticed, which seems strange; For Clock Info Modes used it lists 0 for low, medium, and high... can this be interpreted to mean that it is using Clock Info Settings 0 (Boot) for all three of its power play steppings?

[Boogerlad]

OK...I'm impatient. I stuck the probes in there (man this card is hot!), and my guess was correct. The points below show up as 2.01v on the DMM, so I'm pretty sure they are the measure points for the GDDR. Someone correct my drawing if anything is incorrect, please.



Pic courtesy of Techpowerup


Also, I want to rig the fan to run 100%. Are the red and black fan lead wires the positive and GND? I'm pretty sure they should be, but I don't want to fry the fan. Yellow for sensor, and blue for PWM?

so for vgddr measuring point, one of those yellow points to ground? and for vgddr mod, two whitish points to 100k vr?