8800GT voltage mods

[basco]

thank you very much largon and yiancar !

this is what i did not understand in which direction to read it.
so measure resistance x20 = trimmer.

got it up to 900 linked on ss-hope for 1000mhz+
little bit dissappointed from 55nm performance to 65nm-not much diff-only shader goes higher.
will report back with numbers from hwbot-got my 8600 up to 5700 on ln2 with only 1,65(1,592 load)-hoping for 6,0ghz in mark05.

thanks for your effort and time
jo from austria

[gugle]

oh cool
not sure what the hinix will do if the good chips unlike the quadamas then vmod can be done but i dont know how good they are

Hello, I made the mod (vgpu), would leave the lower voltage of memory? stock is 2.222v

[pest]

I wonder why there was all the talk (from a supposed reputable source too) that 1000 MHz + will damage the card in a short time?

I'm running my 8800GT since Jan08 with 1050MHz Memory
I can run 3DMark06 with 1172MHz, no problems so far



Still i'd like to figure out the Mem-mod for my ECS 8800GT
any help? largon?

Also, what vGPU are you running 24/7? What kind of extra cooling for the PWM are you using? I'd be interested to know since it's lived so long.

Today i added a switch. I benched up to 1.45V, and the card did run until today with 1.36V Idle. Now I'm able to switch while running between 1.1V and 1.36V, saves me 10W
All cooled with an AC S1. The PWM and Memory is cooled with an extra 120mm case-fan.

ignore the dust

[largon]

pest,
Please measure resistance between pin#4 and pin#7.



The vMEM mods are likely identical to those for the reference card (regardless of fact that the chips have different names): vMEM, reverse vMEM.

[pest]

it's a non-reference card, but the memory-controller should be the same?

second problem is that i don't know what memory-chips this card has,
but i guess it's not quimonda because the cooler the better the memory clocks

[largon]

Xbitlabs says that card has Qimonda.
And by the looks of the vMEM power circuit I agree.

[pest]

Thanks for your help and advice
I still think the ram clocks too good for Qimonda (Xbitlabs only managed to clock the mem to 970), but i first try the reverse-mod

[largon]

Do the measurements before soldering...

[pest]

Do the measurements before soldering...

yesterday i wanted to, but i think i have some understanding problems
i am by no means an electronic expert

i don't know the polarity of the pins, so the possibility is very high, that i'm measuring crap

the second problem is, that i have a cheap 9V DMM, and i am afraid that the chip is not capable to stand the voltage

[CryptiK]

I decided to cap mod my 8800GT today I thought if I was going to be running it out of spec, at voltages equal to or above the 3 phase PWM equipped GTS, it may be beneficial to it's lifespan to ensure the mosfets are getting cleaner power, and that the output gpu voltage was also cleaner. I used two 16v 1500uf primary caps, one on each phase before the fets (12v) and two 6.3v 1000uf secondary caps, one on each phase after the fets (vgpu voltage, in my case 1.17v idle, 1.23v load).

Pic of Primary caps side on:


Pics of primary and secondary caps:

yesterday i wanted to, but i think i have some understanding problems
i am by no means an electronic expert

i don't know the polarity of the pins, so the possibility is very high, that i'm measuring crap

the second problem is, that i have a cheap 9V DMM, and i am afraid that the chip is not capable to stand the voltage

When you measure the resistance between the pins largon indicated, you do it with the pc switched off and unplugged. You only measure voltage with the pc running.

So to measure the resistance, turn off and unplug your pc, set your DMM to resistance, put one probe on one pin and the other probe on the other pin and see what it says

[largon]

Those OST branded cans are... suspicious...

[CryptiK]

Yes they sure are lol I will be swapping them tonight with some rubycons I have aside, just put them on in the interim.

[CL3P20]

just curious- what is to be gained from the cap mod exactly.. I am running just over 1.35v on my GS's and wondering if this mod would help..and how it affects GPUv under load.

[CryptiK]

Well not much is really gained, I can reduce the vGPU by 0.02v and remain stable, and the card does not really emit the high pitch squeal it used to under load with high FPS (still emits the noise but is much quieter).

The point of doing it was to smooth the voltage the mosfets were getting and to smooth the voltage the GPU was getting from the mosfets. The 2 phase PWM design is very weak, and for the GTS, which runs 1.16v idle, nvidia saw fit to implement a 3 phase PWM circuit - this tells me they were concerned a 2 phase PWM would be pushed outside its comfortable operating range at that voltage. I added the caps in an attempt to extend the card's lifespan, that's all.

EDIT - I have replaced the OST 1000uf caps with some Sanyo 1000uf caps (WG series, very low ESR), the Rubycons I have are a little large (3300uf) and also physically don't fit under my ram sticks when the card is inserted. The Chemicon KZG caps should be ok but I may replace them with something newer, as they are a few years old but have not been used more than a couple of hundred hours at no more than 60*C, so theoretically they have a lot of life left.

[CL3P20]

Thanks for the info.. I may give it a try if going too much farther with GPUv becomes unstable.. I think I am approaching the GPU limits anyhow at 1028mhz core/ 2214mhz shaders with 1.362v idle and ~1.418v load.

[gugle]

Hello, I made the mod (vgpu), would leave the lower voltage of memory? stock is 2.222v

????????

[largon]

^That question does not compute.
(What exactly are you asking?)

[CryptiK]

I think he's asking if he should do the vmem mod or leave it stock, and that the stock vmem is 2.22v.

If that is correct, personally I'd leave it, 2.22v is more than most cards run stock (mine runs 1.98v).

[largon]

Yep, since his stock voltage is 2.22v the chips cannot be Qimonda, which would benefit from <2.00v.

[gugle]

I need to reduce my vdimm. stock: 2.222v

chips Hynix DDR3 (FP-14) = 700mhz
ECS N8800GT-256MX-F @ 512mb

[largon]

gugle,
See post #1204. Measure the resistances so I can confirm the linked mod (VR values) applies to this card.

[CryptiK]

OK well I removed the Chemicon KZG's on the input side as although they are supposed to be great caps, they can fail suddenly without showing any symptoms first. I don't think the input needs much more filtering, as it already has 2 solid caps per phase. After adding the 1000uf sanyo WG's on the output side, the card is now stable at lower voltage. It used to need 1.23v load for 756/1836, now 1.20v load is sufficient. They have had a positive effect. Hopefully they also help the fets deal with the sudden current draw in light to heavy load transitions.

It has always struck me as strange how on video cards there is very few input filtration caps, and only surface mount ceramic output filtration caps. I was especially surprised to see how few caps of decent size there were on the new GTX260/280 cards.

[CL3P20]

After adding the 1000uf sanyo WG's on the output side, the card is now stable at lower voltage. It used to need 1.23v load for 756/1836, now 1.20v load is sufficient. They have had a positive effect

Thank you for that! I am definitely interested in capping my GS's now.. think I have a old ASUS mobo laying around here somewheres.. gonna check for some suitable caps.

*Thanks again Cryptik

[gugle]

gugle,
See post #1204. Measure the resistances so I can confirm the linked mod (VR values) applies to this card.

pin#4 + gnd = 680 ohm

pin#4 + pin#7 = 0.649 ohm

[largon]

^Weird, cause pin#7 is ground. Getting a reading from pin#7 likely just requires some force with the probe as there's solder flux on the legs.
Anyways, both vMEM mods I linked in post #1204 work on your card too.