Radeon X800 Non-Pro/X800 XL/X850 Voltmods

[Urlyin]

yeah i meant a pic of the actual pcb.. I've seen that in the datasheet ..

My cards water cooled and so hard for me to look at the pcb without having to remove it which takes ages..

Sorry I thought you wanted a pin out .... none of the pics on the first page are close enough?

 

[Unregistered]

yeah i meant a pic of the actual pcb.. I've seen that in the datasheet ..

My cards water cooled and so hard for me to look at the pcb without having to remove it which takes ages..

Bleh.. logged me out and made me double post

 

[Unregistered]

Sorry I thought you wanted a pin out .... none of the pics on the first page are close enough?

Sorry I missed one of the pics, this one hows it quite well:

All of the VID pins seem to jut have solder blobs and go through the board...
Pins 8 & 9 should be grounded at stock (for 1.4v stock gpus) and the rest should be floating.


Stock seems to vary between boards from 1.3 to 1.45v so theres a good chance the VID is set differently and maybe the resitance is different for the feedback? I dont know how ati have done it..

It should be really easy to get 1.6v just by grounding one of the solder blobs (VID2) and that will be exactly 1.6, which is also nicely below the OVP

 

[Kramdra]

I hope this image is helpfull

Ive numbered the VID pins, and coloured the traces and resistors they connect to in red.

The green line shows what a 1.4v card will look like at stock, where as this pic appears to be of a 1.35v card.

This is a table of the VID and the output taken from datasheet (there are more settings from 0-1.3v that I missed off).

Another good thing about this is you can undervolt the GPU if you needed to put the card in a shuttle or somewhere where a much smaller cooler has to be used. Running it at 1v would probably still give ~400mhz on core but maybe half the power output..?


Edit: Resistors for VID are 1k and I think the size is 0603 or 0805 but either would do. They cost about 80p for 100pcs.. http://www.rapidelectronics.co.uk/rkmain.asp?PAGEID=80010&STK_PROD_CODE=72-0797

 

[Tijgert]

Bench time!

As promised, some benches of my OC'd X800XL.

First the hardware setup:
Mem - 205MHz DDR @ 2-2-2-5 1T Trc = 10 + Trfc = 14
CPU - 2660MHz A64 Venice @ 313HTTx8.5
Mobo - MSI K8N Plat. (Neo4 Ultra)
CPU + GPU H2O'd.

Core/mem 500MHz / 600MHz (yeah, not bad eh?)
3DMark 2000 = 26253 marks
3DMark 2001 = 26626 marks
3DMark 2003 = 12490 marks
3DMark 2005 = 5912 marks

The core@510 freezes after a few secs, 505 might work but I like to keep some buffer.
The mem@600 is artifact free, but only at core=500MHZ. At 400MHz core the mem@600 does give some artifacts... dunno why.

I kick the X800XT's ass just looking at the clocks, but comparing scores in 05 I even slap the X800XT PE's behind!

All in all, not bad.

Edit:
At 505/605 (roughly a 1% increase in speeds) I get only 10 marks extra. 10 on top of 5912 is 1.7% extra performance... where did the rest go? I'm guessing a bottleneck somewhere else than the videocard.
Perhaps a 2660MHz Venice just isn't powerfull enough...

 

[Mussels]

my AGP pro got 6,300 at its best, and i've seen an XTPE pass 7,000 at stock.... was AGP on a mean ass PC tho (2GB DDR500, FX 55 etc)

 

[Tijgert]

Wel, obviously I was referring to stock speeds.
6300 on Pro at stock speeds? Kinda hard to believe.

And I got my X800XT PE number from some chart, don't really have that card lying around here.

Oh, and boohoohoo, I'm slower than the most expensive cards... I'll live

 

[funkflix]

my AGP pro got 6,300 at its best, and i've seen an XTPE pass 7,000 at stock.... was AGP on a mean ass PC tho (2GB DDR500, FX 55 etc)

Sorry, but no way to get 7000 with a any stock Ati!

 

[Mussels]

it was one of my friends that got the score, i'll ask him next time i see him.

 

[Urlyin]

Sorry I missed one of the pics, this one hows it quite well:

All of the VID pins seem to jut have solder blobs and go through the board...
Pins 8 & 9 should be grounded at stock (for 1.4v stock gpus) and the rest should be floating.


Stock seems to vary between boards from 1.3 to 1.45v so theres a good chance the VID is set differently and maybe the resitance is different for the feedback? I dont know how ati have done it..

It should be really easy to get 1.6v just by grounding one of the solder blobs (VID2) and that will be exactly 1.6, which is also nicely below the OVP

Very interest Kramdra... I never really paid that much attention the the VID, thinking you needed to place or remove the SMR to obtain the desired voltage. So when you say ground the two solder pads you mean connect the two? Would a conductive pen create the desired effect?

 

[Mussels]

VID? would this be an alternate way of modding?

 

[Urlyin]

VID? would this be an alternate way of modding?

It is if you wish to solder a small SMR .... I don't see the advantage of this over using a VR ... But nice of Kramda to take the time to share his knowledge ...

 

[Unregistered]

Maybe the VID mod will work using a pencil, allowing you to easily dial in the exact voltage. Hopefully, the resistance doesn't need to be exactly 1k.

Now I don't understand why the normal pencil mod doesn't always work since the VR mod is using the same points. It's possible that the solder spots have a bit of coating on it thats preventing the pencil lead from making proper contact.

Anyway, I'm going to try the VID mod with a pencil and see what happens

 

[ViperJohn]

As promised, some benches of my OC'd X800XL.

First the hardware setup:
Mem - 205MHz DDR @ 2-2-2-5 1T Trc = 10 + Trfc = 14
CPU - 2660MHz A64 Venice @ 313HTTx8.5
Mobo - MSI K8N Plat. (Neo4 Ultra)
CPU + GPU H2O'd.

Core/mem 500MHz / 600MHz (yeah, not bad eh?)
3DMark 2000 = 26253 marks
3DMark 2001 = 26626 marks
3DMark 2003 = 12490 marks
3DMark 2005 = 5912 marks

The core@510 freezes after a few secs, 505 might work but I like to keep some buffer.
The mem@600 is artifact free, but only at core=500MHZ. At 400MHz core the mem@600 does give some artifacts... dunno why.

I kick the X800XT's ass just looking at the clocks, but comparing scores in 05 I even slap the X800XT PE's behind!

All in all, not bad.

Edit:
At 505/605 (roughly a 1% increase in speeds) I get only 10 marks extra. 10 on top of 5912 is 1.7% extra performance... where did the rest go? I'm guessing a bottleneck somewhere else than the videocard.
Perhaps a 2660MHz Venice just isn't powerfull enough...

What will it core clock running 25 repeat passes of 3DM01se Nature only with the title screens disabled???

Viper

 

[Unregistered]

!!!SUCCESS!!!.

I changed the voltage from 1.35 to 1.55 by shorting out vid2 with a pencil. I scratched the solder points first with a pin just in case there was any coating on them and then i used a pencil across a few times and that was it.

 

[Unregistered]

P.S. Thanks Kramdra, your idea worked like a charm. This should make gpu volt mods easy as pie

 

[Unregistered]

Ok, with regards to posts 266 and 267, the card that I did the VID mod on is a bba x800xl agp. The stock gpu voltage is 1.35 and it has the missing resistor.

I since closed VID 0 with a pencil the same way as VID 2 and now the voltage reads 1.6v, just as Kramdra's chart indicates.

The resistance of one of the connections measured about 300 ohms though my guess is the value doesn't matter that much (no proof so don't sue me).

Using atitool, my O.C. increased from 460mhz (stock voltage and watercooled) to 479mhz (1.6v).

I'll probably return the card to stock voltage as I feel 19 mhz is too small a gain for .25v. Hopefully other people will have better luck as the mod itself couldn't have been easier to do. In fact, this may prove to be a great mod for people without a multimeter.

Good luck to the next one to try it out,

FLB

 

[Mussels]

can you try that and lower it by say, 500 ohms? if you can make it raise by an amount safe for air cooling, i'll do it myself.

 

[Unregistered]

can you try that and lower it by say, 500 ohms? if you can make it raise by an amount safe for air cooling, i'll do it myself.

Changing the resistance doesn't change the voltage using this method. Depending on which VID I close, the lower voltage options are 1.4v, 1.55v, and 1.6v. I think 1.55v should be safe for aircooling

FLB

 

[Mussels]

sweet, if you could get pics and show how to do this mod.... (edit the pics)

 

[Unregistered]

Warning! this mod will give you 1.6v if your stock voltage is 1.4v. I would sooner rely on the vid configuration to determine my stock voltage than a multimeter as a lot of cheap ones may be inaccurate by .05v

If you have a resistor where the picture (post#255) shows a green line (R1854), Then I wouldn't proceed with this method as I think 1.6v is a little too high for stock aircooling.

So, If your stock voltage is 1.35v then look at post #255 and connect the two points labeled vid 2 (location R1856) with a pencil. I scratched the two solder points with a pin first to make sure I had a good connection with the pencil lead. This will give you 1.55v.

 

[Urlyin]

sweet, if you could get pics and show how to do this mod.... (edit the pics)

Yes... it would be sweet if it works... not sure how the pencil graphite creates enough resistance to create the feedback needed .... If you look at SMR in the pic it's labeled a 102 which is much lower resistance than a pencil can create.... but if it works Great...

 

[Unregistered]

Yes... it would be sweet if it works... not sure how the pencil graphite creates enough resistance to create the feedback needed .... If you look at SMR in the pic it's labeled a 102 which is much higher resistance than a pencil can create.... but if it works Great...

Do you mean to say that a 102 is a much lower resistance that a pencil can create?

 

[Urlyin]

Do you mean to say that a 102 is a much lower resistance that a pencil can create?

The first two (or 3) digits are the first two (or 3) digits of the resistance in ohms, and the third(or 4th) is the number of zeros to follow - the 'multiplier'.
Three Digit Examples & Four Digit Examples
330 is 33 ohms - not 330 ohms
1000 is 100 ohms - not 1000 ohms
221 is 220 ohms
4992 is 49 900 ohms, or 49.9 kohm
683 is 68 000 ohms, or 68 kohm
16234 is 162 000 ohms, or 162 kohm
105 is 1 000 000 ohms, or 1 Mohm
0R56 or R56 is 0.56 ohms
8R2 is 8.2 ohms

Given a 102 labeled SMR... then that would be 1000 ohms or 1k ... not to say that is the required amount to change the voltage... Kramda is correct you just need to jump the connectors .... Mussels you should be able to use the diagram by Kramda to vmod your card, in other words it should work

 

[W1zzard]

updated the first post with images for kramdra's vid mod.

if anybody needs any more high-res pics of any other voltmod articles i did, let me know.