ATI 4850/4870 voltmod thread

[dejanh]

OK, I've got a problem measuring, probably cos of my incompetence, but wanted to recheck with you guys.

I did use the above URL, and picture that is shown there. But I couldn't get any numbers from my multimeter, though it could be that I've missed the point somehow, even though I've tried 5-6x..

So let's take it step by step, as I don't use MM usually, and I've borrowed this one just for the occasion

1) I take multimeter and turn it on
2) I have a switch and a dial to set up the desired range
a) little switch that selects amps/ohms/and two types of voltages. One is V (that's DC right? or not?) and the other is V~ (AC?). I'm not certain which of the two is right (that's why I've included pictures, I hope they are shown right in post), as I'm not sure if GPU voltage is AC or DC.. Though I was just struck with the VDDC markings, but it could mean anything as I still don't know what MVDDQ means either except that it's some memory voltage So I guess it's DC, but tell me anyway, is it this one or this one ?
b) next I have a dial that's similar to this one ; it lets me select 2/20/200 for whatever I've selected on the switch (ohms, amps, volts). Should it be better to select 2V or 20V, since VGPU is <2V actually, I guess 2V is better and more precise, right?
3) I take those black and red umm "cables", and I put one on the VDDC measurement point on the VGA card, and other one where? As much as I know, it's just important that circle is closed, so I've just put the other side on the PC case.. It worked fine when I've tried to measure current on peripheral/molex plugs, showed 5V and 12V no problem.. But if VGPU is DC than what? And I guess than there is difference between using red or black one.. So a bit more info in here as well..

There are few things I could have missed, I'm no longer sure, but maybe I've switched to wrong voltage type (step 2-a). Also, maybe my PC case is not good for closing the circle when measuring VDDC on GPU (step 3), or i've just placed wrong red/black cable on the measurment point. Also, I could have missed the measurment point completely, so I didn't have good enough contact. Anyway, please, step-by-step instructions would be real nice, so I can do some measurments today or tomorrow when I should also have IR thermometer and power consumption meter, so I can check what are the effects of voltage/clock changes in BIOS, and how will it all affect stock cooler vs aftermarket cooler..

Thanks in advance

DC, 2V, red lead on voltage measuring point, black lead on ground (e.g., side of your case).

[LuxZg]

DC, 2V, red lead on voltage measuring point, black lead on ground (e.g., side of your case).

Thanks man, I just figured it out few minutes ago. Could you tell me if I got right measurments before I go to restart and try once more on a "clean" system (I've had ATT running while measuring, maybe that scewed results a bit)

Ok, checked:
stock clocks 750/900
1.281V 2D idle, 1,294-1,298V under Crysis bench

overclocked 790/1100
1.284V 2D idle, 1,297-1,300V under Crysis bench

Weird how I got bit different than when I had ATT running But sounds correct enough.. Now time to do some modding
Cya!

[dejanh]

Thanks man, I just figured it out few minutes ago. Could you tell me if I got right measurments before I go to restart and try once more on a "clean" system (I've had ATT running while measuring, maybe that scewed results a bit)

Ok, checked:
stock clocks 750/900
1.281V 2D idle, 1,294-1,298V under Crysis bench

overclocked 790/1100
1.284V 2D idle, 1,297-1,300V under Crysis bench

Weird how I got bit different than when I had ATT running But sounds correct enough.. Now time to do some modding
Cya!

Yeah, those voltages sound about right.

[stingush]

here's my E7200+4850 :
AC S1 38 idle 52 full load, 1.33VGPU + 2.1V for DDR3
E7200 should work on lower V but i didn't have to much time so i made it the safe way...

[dejanh]

Way to hide the desktop pr0n

j/k, nice clocks

[Barr3l Rid3r]

here's my E7200+4850 :
AC S1 38 idle 52 full load, 1.33VGPU + 2.1V for DDR3
E7200 should work on lower V but i didn't have to much time so i made it the safe way...

Why? I made 15500 using a E8400 @ 4.5Ghz 530MHz FSB and much highr clocks in mine 4850 (900/2200) and u have about the same score?

what driver or tweaker are u using? Is that all caused by Vista 64?

[stingush]

i dont know, im using Cat 8.7 and thats it, simple overdrive OC with good cooling on gpu and memories of the card, thats all. guess it is vista x64..
also, my winxp sp2 is full of **** so i guess if i had a clean installation of the windows i whould get something like 157XX or so...

[LuxZg]

Info for everyone!

LINK TO THE POST with some new BIOS from ATi, with some interesting stuff in it. I'm at work so can't test it, but read the post text and it will sound interesting. Specialy if this is really from ATI and not from Diamond (guy did say ATI explicitly, so should be official).

It has unlocked CCC overdrive, and seems they did something to voltages as well. You'll get exe file that auto-flashes the card, but you can extract it (I've done it with PowerArchiver, but probably WinRAR will work too), and you'll get the files. "4870-128Kb512MBOC795-1100MHz.sb" can be opened in RBE no problem and shows this:
BIOS DATE: 08/13/08 13:01
VERSION: 113-BA0701-100
ATOMBIOS VERSION: ATOMBIOSBK-ATI VER011.003.000.001.029254
500/1100/1,276V 2D
795/1100/1,276V 3D

Note - higher voltage! As soon as I get home I'll flash it, try it, and see if it really ups the volts, and if you can use lower volts as well.

Also, 990/1250 is upper OC limit in CCC.

This should be official ATI driver for boards based on reference design, so start flashing ppl!

Btw, I have posted or will be posting this in other threads too, as it is VERY interesting info, and I guess as many people know about it and test it - the better

Cya all!

EDIT: note - this is for 4870 cards

[deathman20]

In the newer version of RBE it has the option to increase the clocks for CCC as well if nto mistaken.

[LuxZg]

It does, but no BIOS so far managed to push voltages up for 4870 cards.

I'm home now and I've tested few things.

By flashing this bios you really do get more juice, even if by a small ammount. Earlier under load I've measured 1,307-1,312 V
(overclocked to 790/1100) and now I get 1,331-1,334 V on 795/1100.. so this is 0,02V more. And no it's not cos of additional 5 MHz, since even 50MHz hardly pushed it for 0,002V

Other than this, I've tried if this BIOS allows lowering voltages, but as much as I've tested - nothing happens. I've tried by new Ati tray tools, by RBE low voltage tweak, by entering values manualy in bios - nothing helped.

I did NOT try rising voltage in BIOS, but with ATT it apparently doesn't move. I believe this BIOS has same voltage-changing limitations that the reference one had, only now it's fixed to 1,276V instead of earlier 1,263V. You can try it yourselves, and good luck

I'll keep playing with the low-volt BIOS cos I'm mostly limited by the CPU anyway...

What we can expect now, is that someone will compare original BIOS and this one, and maybe will figure out WHICH bits exactly change voltages either to higher or lower values.. I hope RBE will be able to do it in a release or two

[Dirton]

You are saying you are not failing due to over heating at 1.35V, but you are running your GPU at 1.263V which is the default voltage (not 1.283V). Try actually running FurMark with 1.35V and high clocks. Otherwise, good observations, there is a definite link between maximum core and maximum memory

Try and measure it for yourself under load, it's 1.283v.

I can actually hit pretty high at 1.35v with water, but unfortunately when I try OC memory a bit it goes down fast. For example I can pass Vantage with maxed memory only on 890/1200 (gpu/mem), and with no vmod and stock cooling I can do it on 875/1200. Don't seem normal to me.
What I was trying to say, there is no improvement after upping the voltage and that's abnormal. Well there is if I OC only GPU in furmark:
default voltage - 860MHz
1.35v - 940MHz
1.4V - 965MHz
And another thin, when I try to pass Vantage on default voltage with maxed MEM clock, I actually can do it 15MHz higher then furmark can allow. With 1.35v it already becomes 50MHz lower! Now, that's what worries me.

In the newer version of RBE it has the option to increase the clocks for CCC as well if nto mistaken.

Well, it does, but it isn't working.

[LuxZg]

Well, it's working now, cos RBE 1.14 is out

[0maro]

Hello everyone,
there is my result.
Gigabyte HD4850 + fv1000led(i'm planning to mod that sink).
Core-1.33V/812mhz;
Mem-2.25V/1125mhz.
Stable.



My plan is to get ~900mhz on core .. :\

[Barr3l Rid3r]

Hello everyone,
there is my result.
Gigabyte HD4850 + fv1000led(i'm planning to mod that sink).
Core-1.33V/812mhz;
Mem-2.25V/1125mhz.
Stable.



My plan is to get ~900mhz on core .. :\

Don't even think to try that whith this terrible cooler.

[Barr3l Rid3r]

[eddieobscurant]

nice gpu clocks barr3l rid3r , could you tell us what's your gpu voltage? Are you 100% stable there or just for beches. What's you max stable?

[dejanh]

And the saga of failed non-reference 4870s continues (definitely worth a read as what they are seeing is an ongoing issue with all 4870, especially with after market cooling)...

http://en.expreview.com/2008/08/23/p...-issues-check/

[Barr3l Rid3r]

nice gpu clocks barr3l rid3r , could you tell us what's your gpu voltage? Are you 100% stable there or just for beches. What's you max stable?

850 @ 1.34v fora daily use

[Rick Flair]

For 4870 users who want to use fixed resistors for the vmem mods, 4.7K Ohm resistor will yield 1.61v.

[LuxZg]

@dejanh and powercolor card
Well, I've had no problems with my 4870 even on 800/1100 and stock cooler, I just modified BIOS fan tables so I have low fan in 2D and high in 3D and that's it.. And I stay away from Furmark, cos if it runs all games just fine for hours and under 90C VRMs, why the hell would I want to run furmark for over 126C after 30 seconds?
Now I've mounted Accelero S1, and I have to say that manufacturers will have a hard time keeping this card cool with inferior cooling as this Powercolor had. I had S1 and two 120mm fans, and it was still no better under very high load than stock cooler was on 60% fans (except that it was quiet). So I had to mount a higher powered 120mm fan which is not noisy but can be heared even when it's not complete silence in the room, and only now do I have very good temps when not overclocked, and just about the same as stock @60% fan when OCed to 800/1100 and under heay load for an hour or so..

So I don't think that any manufacturer will get this one right. And the reason is not even in GPU speeds but in memory. This GDDR5 draws a lot of current, and it's the number one reason why both VRMs and even GPU heat up. Cos when I clock memory up, "MEMIO" sensor on GPU goes sky-high even with a good cooler like S1+dual fans, and that heats up whole GPU chip a bit.. Also, amps go up a lot, and that heats VRMs. So they can clock the GPUs but I wouldn't recommend anyone clocking memory after everything that I've tried. One more point for that is - this card is really not held back by memory. I'd say it's held by CPU (for anything not clocked over 3.5GHz), than GPU, and only than memory..

So people - for 24/7, just don't clock memory.. no need, and just a bunch of potential problems. Potential, but almost certain..

[dejanh]

@dejanh and powercolor card
Well, I've had no problems with my 4870 even on 800/1100 and stock cooler, I just modified BIOS fan tables so I have low fan in 2D and high in 3D and that's it.. And I stay away from Furmark, cos if it runs all games just fine for hours and under 90C VRMs, why the hell would I want to run furmark for over 126C after 30 seconds?
Now I've mounted Accelero S1, and I have to say that manufacturers will have a hard time keeping this card cool with inferior cooling as this Powercolor had. I had S1 and two 120mm fans, and it was still no better under very high load than stock cooler was on 60% fans (except that it was quiet). So I had to mount a higher powered 120mm fan which is not noisy but can be heared even when it's not complete silence in the room, and only now do I have very good temps when not overclocked, and just about the same as stock @60% fan when OCed to 800/1100 and under heay load for an hour or so..

So I don't think that any manufacturer will get this one right. And the reason is not even in GPU speeds but in memory. This GDDR5 draws a lot of current, and it's the number one reason why both VRMs and even GPU heat up. Cos when I clock memory up, "MEMIO" sensor on GPU goes sky-high even with a good cooler like S1+dual fans, and that heats up whole GPU chip a bit.. Also, amps go up a lot, and that heats VRMs. So they can clock the GPUs but I wouldn't recommend anyone clocking memory after everything that I've tried. One more point for that is - this card is really not held back by memory. I'd say it's held by CPU (for anything not clocked over 3.5GHz), than GPU, and only than memory..

So people - for 24/7, just don't clock memory.. no need, and just a bunch of potential problems. Potential, but almost certain..

Agree completely. It is not the stock cards that tend to be problematic, it is the "factory OC" ones that do. Clearly the manufacturers do not really have a good idea of the cooling needed on these cards, especially the VRMs. That's basically what kills them all and exactly what I stated. Indeed, I found that it is the two memory VRMs that get ultra-hot. GPU VRMs get hot as well, but not as much as the memory ones. Keeping memory lower is a good idea. The memory is not a bottleneck for this card and only real benefit from OC comes when you OC the core. The minimum CPU to not cause bottleneck on this card is about 3.6GHz Core 2 (AMDs just are not going to do this ATM). The only thing I disagree on is that in my opinion the card should be able to run anything (including FurMark) without any risk of the card experiencing problems or worse yet frying. Maybe ATI engineering team and their manufacturers should use FurMark to stress their cards

[LuxZg]

@ Furmark testing in ATi

Ok, I agree there shouldn't be any problems, with whichever software.. But ATi solved it, right? Tried Cat's 8.8? They halve the Furmark score, and that's it

[dejanh]

@ Furmark testing in ATi

Ok, I agree there shouldn't be any problems, with whichever software.. But ATi solved it, right? Tried Cat's 8.8? They halve the Furmark score, and that's it

New ATI motto - "If you can't fix it, break it!"

[DJSUB]

is there anybody know where to adjust the 2d settings in ATi 4850? i already tried adjusting almost all the values in RBE but it seems no improvement in 2dbench in pcmark05

thanks in advance

[Dirton]

You can do it with RBE or ATT automatic 3D/2D profiles, if your goal is to lower clocks. If now, then you can simple edit CCC profile settings manually.