ECS Xabre 200 Voltage mods thread

Hi all

As I was bored I though id give this a go
Im supprised nobody else has got there before me TBH.

http://www.anachip.com/downloads/dat...ear/AP1084.pdf

Ok we have two Vregs for the VGPU in paralell. They are the simplest type of Vreg... put a voltage in and get one out with one pin for setting it up via a resistor devider.

From looking at the card both Vregs are on the same resistor devider thats in the pic.

I havent done this mod myself yet as I havent even tested the card :P but its pretty simple.

This first pic shows the main components...
If you put a VR over the lower resistor in the devider you should get more voltage.
I think I will just replace this resistor with a VR.

The second pic shows the different voltage planes.. should anybody want to add some capacitance.. which may be a benefit.

Voltage planes on the card.

EDIT .... changed 5v to 3.3v

Ill take a look at the memory tonight.. although I think ill have to power the card up to work this one out.

Its possible that where I have put 5v it may be 12v... I havent powered up the card to check yet as I dont have my number2 box working atm.

EDIT - Turns out its 3.3v

Think a 10k variable resistor will suffice? If so, I'll try it when I get home in a couple hours.

BTW, :toast: for looking into this. ;)

On my card the lower resistor measures only 60ohms.... but i think the resistor is 98ohms.

so 10k may have no effect, but its very unlikely to cause an overvolt...

Can you check the resistances on your own card to confirm my readings 60k seems a bit low to me..?

Im looking into the Vmem mod atm.. it looks simple and complex all at the same time.

VMEM mod

This is the PDF for the samsung chips on the card.

http://www.samsung.com/Products/Semi...k4d64163he.pdf

The memory is powered by 3.3v and I/O voltage is 2.5.
This is best explained page 10 of the PDF

We have 3 voltages to track... VDD which is 3.3v, VDDQ at 2.5 and Vref at 50%of VDDQ (1.25v)

http://www.xtremesystems.org/forums/...threadid=22342

pics of the card by saaya.. my camera isnt working right ATM I have got some much clearer pics.. ill post them when I can.

Quote:

---

Originally posted by Holst
On my card the lower resistor measures only 60ohms.... but i think the resistor is 98ohms.

so 10k may have no effect, but its very unlikely to cause an overvolt...

Can you check the resistances on your own card to confirm my readings 60k seems a bit low to me..?

Im looking into the Vmem mod atm.. it looks simple and complex all at the same time.

---

So it should be around 98ohms right? I'll unsolder one side of it to check. And if we are talking about this type of resistance, what value variable resistor should I start with? I also have a 1k 20-turn pot that I think would work better in this case... that wouldn't be too low, you think?

Ok, ive poked the card a bit more...

The memory runs straight from the 3.3v line as expected.

The VGPU is also running from the 3.3v line droped down to 2v via the mosfets... the mod will be the same however the max voltage will be lower... some people may want more than 2.5v (my estimate of the max voltage avalable)

The mosfet marked 882 on the top right of the card produces VDDQ and im currently trying to work out how it works... I cant find a PDF for it ATM but ill keep looking.

Im not sure how Vref is produced yet...

If your interested the pin out for the AGP connector is below.. page 231.
ftp://download.intel.com/technology/...oads/agp20.pdf

1k might be ok.. still might get a bit scary towards the end of travel... I have a better VR selection here soI think ill just swap the resistor with a 200ohm VR that way ill have much more controll.

But I think 1k will work... just be carefull.

I think I have a VDDQ mod worked out but I need a pic of the card to show it (saaya's pic has a cap in the way)

Just found a 500ohm VR in my "bag-o-goodies". ;)

BTW, in-circuit I'm measuring 68.9ohms for R2.

Not sure of this resistors real value because... I dropped it. ;)

BUT, I just added my 500ohm VR and then set it so the in-circuit resistance was 68.9ohms.

Good news... my card still works with 2.01v.

GREAT news... I've adjusted it to 2.05v with just an 1/8th turn of my VR.

:toast: Holst!

Not sure if this helps overclocking yet... but I'll try a little later.

Nice one m8, im glad it works... post results in the competition thread as soon as you can.

If you replace R223 in a similar maner you should be able to ajust VDDQ (should be 2.5v stock)
VDDQ can be measured as in the pic below.

You "may" get a higher memory clock by just increasing VDDQ.. however you might also need a big boost in the 3.3v line to get any real gains. Carefull you dont kill your card memory vmods are somewhat dangerous.

Im also still unsure as to how Vref works.. ill take a further look now.

Another option may be to cut the 3.3v line to the memory and force in another voltage... however this is a much more complex mod that would need testing... this is my eventual plan (take 3.3 to say 4v) but it will take some practical tests.

Just increased the GPU to 2.1v and then measured R224 in-circuit... it measures 66.1ohms. I guess extremely small changes in resistance affect GPU voltage modestly.

R224=68.9ohms --> GPU=2.01v
R224=66.1ohms --> GPU=2.10v

I'm not going to be able to mess with the card anymore tonight... but hopefully either someone else will tomorrow or I plan to play around Sunday.

At any rate, GREAT WORK!!!

Ive taken a further look at Vref.

It "seems" to be produced through a the VDDQ line via a simple resistor... so it should "sort of" track with VDDQ... to get it 100% right may mean allot of tricky soldering and messing around...

However ill have to read samsungs PDF's allot further before I can say for sure how it works as ive never done this sort of thing before.. memory and associated voltages are somewhat new teratory.

I also have a "possible" solution for VDD on the memory.. however im keeping it under my hat atm..

Holst and Jeff:)

This is great work you guys are putting together. I have only time to bench the card yesterday for a few hours and only checked voltages. Your due diligence is so greatly appreciated.

BTW....my initial results were quite remarkable, especially where the ram is concerned. Overclocking it by itself..and I have the TC50 Samsung, I attained 310Mhz throughout all 3Dmark tests with NO artifacts. 320Mhz made it through all but nature where it froze.

The GPU is like all others, and despite an MCW50T attached fed with -25c coolant 265Mhz max was attainable with only marginal artifacting.

In combination, I could run 265/300. I'll be aquiring some lower resisance VR's and attemping the volt mods myself soon.

Keep up the great job !!

Randi:D

Heres a pic of the finished Vgpu mod. I carefully removed the lower resistor in the divider and soldered in a fan pigtail. so far so good, as Jeff getting approximately 2v @ 68 Ohms. It's 2:30am as I write this, so further testing will occur after some sleep:)

This is a major pic size, but thought it would assist the members who have not attempted this type of mod, with a little higher detail.

Randi:D

Another angle on things:

What VR did you use????

Id recomend using somehting small.

around 200ohms will be fine.. although jeff used 1k.

I ended up using a 500ohm variable... I'd recommend the 200ohm one like Holst suggests.

To make soldering easier for people, in Randi's second picture, you can solder the top solder point to the Vregs bottom leg.

And you can solder the bottom solder point to any ground.

It's a bit of a pain to get it how Randi has it and unless your pretty good with the soldering iron you may give yourself a headache trying to copy him. ;)

Jeff, can you clearly identify:

Top solder poin
Vregs bottom leg
Bottom solder point

Sorry but I am an ignorant in electronics, my english is not perfect and the soldering iron bites me....:D :D :D

Thanks

This is pretty much the final mod.