Radeon HD 4850 Voltmods

[Wile E]

My card cuts out if it hits over 1.4V under load, regardless of temps or clocks.

But, if you are 100% positive the card never goes above 1.4V, then I would make sure your vrm cooling is up to the task.

[PVTCaboose1337]

Oh looks like my pic 404'd... Anyway... Ohm's law is:

V = IR

V is volts, I is amps, R is resistance!

EDIT: Here is my question thought, what I REALLY care about. Someone do the math... Clearly we want both voltage and resistance to change, that means we would have to make amperage a constant?

[theorw]

Originally Posted by PVTCaboose1337

Oh looks like my pic 404'd... Anyway... Ohm's law is:

V = IR

V is volts, I is amps, R is resistance!

EDIT: Here is my question thought, what I REALLY care about. Someone do the math... Clearly we want both voltage and resistance to change, that means we would have to make amperage a constant?

Well isnt P=V*I ??
I dont think that the amperage is a constant...Wattage (P) increases so amperage increases as well.

[theorw]

Originally Posted by Wile E

My card cuts out if it hits over 1.4V under load, regardless of temps or clocks.

But, if you are 100% positive the card never goes above 1.4V, then I would make sure your vrm cooling is up to the task.

I monitor the card having the DMM plugged in all the time and the volts never go over 1.38.
In addition, my vrm have sinks on both chokes and mosfets and the musashi blows over them...
So it might be age or the card reaching its limit on the vtm capabilities...

[PVTCaboose1337]

Originally Posted by theorw

Well isnt P=V*I ??
I dont think that the amperage is a constant...Wattage (P) increases so amperage increases as well.

But then again do we actually know wattage or amperage? Actually we might know wattage which is why that equation might work to get the amperage and then plug it into my equation posted above.

[theorw]

Originally Posted by PVTCaboose1337

But then again do we actually know wattage or amperage? Actually we might know wattage which is why that equation might work to get the amperage and then plug it into my equation posted above.

Well i think 4850 has 110 peak wattage so it goes like this:P=V*I=>110=1,158*I=94,...AMPS.
Well its odd isnt it?I dont think that the current is 94 amps
Am i getting the default voltage wrong or what???

havent used these equations since highschool....

[PVTCaboose1337]

94 amps seems VERY high. I will see if I can find the technical data sheet on the subject.

[Wile E]

The overall power draw of the card on the psu occurs at 12V. So, 110W on a 4850 = 9.16A total power draw.

[belial13]

so i got a sapphire 4850 for some time now, one of the firsts and strapped S1 on it with dual fans

today i finally tried the pencil mod with a B2 pencil and NOTHING, not .01 volts of "improvement" , i am supposed to shade that middle one right ? and one thing i noticed, i got the uP6201BQ chip but its not VLP824 but some other (will check later) does that matter ?

[Vhozard]

I did the first (easiest) pencil mod and it worked out great!
I came from 670 / 1065 and ended with 770 / 1100 stable.
My load temps did increase, from about 55 to 74.

Before:



After:

[theorw]

Very nice job mate!!!What volts do u have @ these clocks?
Take it for someone that v modded to of them in xfire(me)
@1,35 core/2,35 mem they can do 820ish/1200 24/7!
Just have good cooler and sinks on memory and VRMs.
I had 2 musashi...

[luke05]

Deleted, I've changed my video card!

[overclocking101]

Originally Posted by Wile E

Be careful with the mem voltage. My experience with it shows that GDDR3 is pretty sensitive. I quit doing vmem mods for 24/7 use. I do it on benching cards only.

And I vmodded my 4850 30 minutes after I got it. Popped it in to make sure it wasn't DOA, took it back out, fired up the soldering iron, did the vgpu mod, threw a water block on it, and threw it back in. lol.

1.4v got me to 900Mhz bench stable, probably around 870 game stable.

well you sure were right about this! fried my memory controller big time! I can still get the card to post and boot windows but major artifacts as the memory isnt getting correct constant volatge lol. i had it running 790 core and 1150 mem and trimmed up the mem voltage a bit more looked down an was watching sparks come off the mem controller shut down and boom dead controller oh well you live you learn. I got another 4850 coming this time i wont vmod the memory just the core lol

[luke05]

Hi,

this is my HIS 4850 IceQ TurboX - 512Mb:




I think it's non a reference one, so... I'l like to do just a simple pencil mod for the VGPU,
can you help me, please?

It seems that it's not possibile to do the "first" pencil-VMOD, because there is no resistor there.

The second one, for more that 1,4V for VGPU, is possible, according to you or not?

[Wile E]

Originally Posted by overclocking101

well you sure were right about this! fried my memory controller big time! I can still get the card to post and boot windows but major artifacts as the memory isnt getting correct constant volatge lol. i had it running 790 core and 1150 mem and trimmed up the mem voltage a bit more looked down an was watching sparks come off the mem controller shut down and boom dead controller oh well you live you learn. I got another 4850 coming this time i wont vmod the memory just the core lol

Yeah, I don't know if it's just my luck or what, but I have a 100% failure rate on vmem modded cards. And we aren't talking crazy voltages, I'm talking going from a stock 2.0 to 2.05 or 2.10 with additional cooling on the memory.

[pita]

Hello,i am new here and i wanted to ask how low can i put the resistance on gpu mem,and where i could measure gpu memory voltage?My card is sapphire 4850 dual slot fansink with accelero s1 on it.http://img341.imageshack.us/i/pict00...ct0068a.th.jpgThe picture.

[theorw]

Originally Posted by luke05

Hi,

this is my HIS 4850 IceQ TurboX - 512Mb:

http://www.pctunerup.com/up/results/...4Turbox512.jpg http://www.pctunerup.com/up/results/...ox512_Vmod.jpg


I think it's non a reference one, so... I'l like to do just a simple pencil mod for the VGPU,
can you help me, please?

It seems that it's not possibile to do the "first" pencil-VMOD, because there is no resistor there.

The second one, for more that 1,4V for VGPU, is possible, according to you or not?

Having modded 2 4850s,reference and not i a pretty sure its reference design so look at the forums start for more info on pencil orhard mod!

[pita]

Originally Posted by pita

Hello,i am new here and i wanted to ask how low can i put the resistance on gpu mem,and where i could measure gpu memory voltage?My card is sapphire 4850 dual slot fansink with accelero s1 on it.http://img341.imageshack.us/i/pict00...ct0068a.th.jpgThe picture.

bump

[arroyo]

Hello,

I want to make v-mod on the reference HD4850 simplest way possible. I do not trust my soldering skills and I do not like pencil mods, because they change their properties after some time. However, I ran an idea:

What do you think about this:
1. I change BIOS voltage values using RBE 4 units down (instead of 1.158V will be 1.006V, etc.)
2. I would flash the card with the modified bios.
3. I would paint a resistor with conductive paint using method similar to pencil mod (method for below 1.4V)
4. I would OC and change the voltage in bios.

What I should get:
Since the pencil mod on the resistor provides a maximum of 1.4V under stress, I will reduce this value in the BIOS to 1.24V. I could peacefully (without the use of potentiometer) flash card to voltage that I like (maximum of 4 units higher, so 1.28V, 1.32V, 1.36V and 1.4V). Resistance will not change after a month, because mod will be executed by paint instead of graphite pencil.

What am I afraid:
- That the short-circuit resistor hurt anything.
- That a maximum of 1.4V at the method of the resistor is fake, and indeed it is a much higher voltage
- That despite the settings in the BIOS, the card will receive the maximum voltage (> 1.4V)
- That no one never tested it and it is madness.

Can anyone allay my fears before I burn my HD4850

[Necrofire]

Originally Posted by PVTCaboose1337

94 amps seems VERY high. I will see if I can find the technical data sheet on the subject.

Well, 110W would be the whole card, not just the core.
But, at 110W, the card would be pulling ~9.2Amps at 12V. This would come out to ~94A if everything on the card worked on 1.158Volts. So the core alone draws somewhat less than 94 amps total, but it would still be a hefty number given how relatively low the voltage is.

Think about it this way:
Given a cpu says it's TDP is 125W. If the voltage is 1.4, then it would be using ~90 amps.
I bet those traces would start getting warm at that amperage.

[Necrofire]

Originally Posted by arroyo

Hello,

I want to make v-mod on the reference HD4850 simplest way possible. I do not trust my soldering skills and I do not like pencil mods, because they change their properties after some time. However, I ran an idea:

What do you think about this:
1. I change BIOS voltage values using RBE 4 units down (instead of 1.158V will be 1.006V, etc.)
2. I would flash the card with the modified bios.
3. I would paint a resistor with conductive paint using method similar to pencil mod (method for below 1.4V)
4. I would OC and change the voltage in bios.

What I should get:
Since the pencil mod on the resistor provides a maximum of 1.4V under stress, I will reduce this value in the BIOS to 1.24V. I could peacefully (without the use of potentiometer) flash card to voltage that I like (maximum of 4 units higher, so 1.28V, 1.32V, 1.36V and 1.4V). Resistance will not change after a month, because mod will be executed by paint instead of graphite pencil.

What am I afraid:
- That the short-circuit resistor hurt anything.
- That a maximum of 1.4V at the method of the resistor is fake, and indeed it is a much higher voltage
- That despite the settings in the BIOS, the card will receive the maximum voltage (> 1.4V)
- That no one never tested it and it is madness.

Can anyone allay my fears before I burn my HD4850

I don't see why it wouldn't work, but one word of warning:
Conductive paint has a much lower resistivity than graphite. You basically risk just shorting the whole resistor with even a very small amount of paint.

[theorw]

Originally Posted by Necrofire

Well, 110W would be the whole card, not just the core.
But, at 110W, the card would be pulling ~9.2Amps at 12V. This would come out to ~94A if everything on the card worked on 1.158Volts. So the core alone draws somewhat less than 94 amps total, but it would still be a hefty number given how relatively low the voltage is.

Think about it this way:
Given a cpu says it's TDP is 125W. If the voltage is 1.4, then it would be using ~90 amps.
I bet those traces would start getting warm at that amperage.

dude u are making a mistake...
The card uses 12v that is then transformed into 1,158...so 12v are the ones to use in the P=VI formula....so the AMPS are 9,1A...94A cannot be supplied by a PSU..

[Vhozard]

UPDATE: I freakin' trashed my card!

It all started good: increased GPU clocks and memory clocks, but also a temperature raise (not too big luckily).

But then one particular game started crashing: World in Conflict would crash after 10~20 minutes (random), making my screen filled with blue/red/gray vertical lines.
Maximum temperatures were only 74 degrees under load, but it couldn't have been anything else than my HD4850.
So I tried to lower the clocks to stock, but that didn't help.
I even lowered them to 500/750, but no deal. I went from to

Now I've cleaned my card with rubbing alcohol and it still crashed.
Only after cleaning up the pencil mods and lowering the clocks to 690/995 it doesn't crash.
FurMark makes my card 50 degrees tops (after 1 hour Xtreme Burning) and shows no artifacts.

Maybe I fucked up when I applied the memory pencil mod? Or was the GPU mod the killer...
FurMark (1.8.0) can't crash my card with all settings on (including Xtreme Burning Mode) while showing 99~100% load in GPU-Z, but World in Conflict can?

PS You can rule out memory/cpu/psu problems, it's definitly my HD4850

Some more pics of cooling (Accelero S1):

[Wile E]

The artifacts you mention suggest memory. I highly suggest NOT doing vmem mods. GDDR seems much more sensitive to voltage than standard DDR memory types. Every card I have ever vmem modded, even by as little as going from 2.0 to 2.05V has resulted in degradation for me.

Do the vgpu mod, and leave the mem volts alone.

[Vhozard]

Yes, that's what I've learned too. Memory mod probably screwed it up.