View Single Post
Old 03-22-2012, 07:44 PM
TiN's Avatar
TiN TiN is offline
Volt-Nut :)
Join Date: Jan 2010
Location: Taipei
Posts: 1,458
Send a message via ICQ to TiN Send a message via Skype™ to TiN
Talking Uncorking KELPER - practical guide for beginners.

During few days of our experience with overclocking EVGA GTX 680 cards with Vince I did some basic modifications which
I can share with fellow overclockers who still remember how to hold soldering iron in hand.

Most of modifications are common between different GPU's so those who bench with extreme should be very familiar with
measuring resistance, calculating proper VR value, checking if any shorts, finding ground and using DMM. So I would not
go crazy details about every small thingy possible, but will give some notes and ideas how to deal with GTX 680 to
reach max OC.

Those who interested in specs, card features and comparsion analysis can easily find that info in reviews and
articles about new-gen 28nm NVIDIA GPU, so here we focus straight on real bussines with soldering.

First let's take a brief look on VRM section of card.

We see uncommon in PC's staggered MiniFit JR with two 6pin power inputs for +12V, and empty space for regular 6pin.
GTX 680 is have lower TDP than monsters like GTX 480 or 580, so 6-pin is usually enough for most of OC conditions, even on water
or subzero. Memory VRM is similar to one used since GTX 480, two phase with discrete FETs.

Main VRM is done to support 5 phase, but in final reference boards was defeatured to 4 phase, what's enough for aircooling.
Also GTX 680 uses input power sensing curcuitry to calculate input power and adjust it's speed and voltage to meet desired
performance and power limits.

On back we see new idea to have main GPU PWM as soldered-on module with related passives. This was probably used for debugging
and testing VRM with different PWM controllers. Also can see one more sensor for input power. On bottom part we can see six empty
footprints for tantalum capacitors for better decoupling of GPU power plane. Luckily for external power users, specially EVGA EPOWER owners this GTX 680 design is really friendly modification-wise. GPU power copper plane is located on outside surface of PCB, so you can easily shhrrr-shhrrr protective solder mask and solder crazy AWG10 wiring directly, without risk to broke capacitor pads or tiny islands on PCB. Also GTX 680 have two PLL LDO's, each of those similar to ones on GTX 580 and 480.

From my brief experience kepler does not need crazy voltage and current to run fast clockspeeds, we were able to hit 1350-1400MHz on
aircooling with just VID mod to 1.3V on GPU. So I will describe modification of onboard power first, and then we will have some talk about external power module, on my example EVGA EPOWER 8-phase or 10-phase retail boards.

GPU Voltage modification

Main PWM for GPU is Richtek RT8802A. Luckily for us, full datasheet on this PWM controller is available public on official vendor web-site. It's quite simple 5-phase controller with VID control. It have one drawback tho,
OVP is not adjustable and fixed internally to offset from VID setting, so to prevent triggering protection during heavy loading/overvolting good way is to use VID mod. I took multi-position DIP switch, and connected one side of it's poles
to ground, and another side to VID pins. To be able to toggle VID i used pullup resistors 4.7kohm to +5V, so when switch
position open - i get logical "1" on VID signal, if position shorted - logical "0". Then task simplifies to just finding
needed voltage in VRD11 table in datasheet above and setting proper levels for switch.

NVIDIA limited voltage range to max 1.21V by hardware, having MSB VID6 and VID7 tied to 1 and 0 by hardware (direct connect to power and ground by trace). So if you want VID more than 1.21V than need to cut VID6 connection to VCC and manually connect to GND. Because VID7 already tied to GND I just shorted those two pins together with piece of thin wire.

So total modification can be look like this:

This worked ok, but when I tried to push 1.5V on one of cards FET in one of phases failed with bad smell, so be careful with going high volts. As usual - all modifications provided AS IS for our own risk.

Also not hurt to populate back solid caps which was supported by reference design by initial idea.

Memory voltage modification

Memory is simple to modify, it's Richtek 2-phase PWM without any advanced controls. So just usual VR's for feedback adjustment (marked red on photo above). Also if you want to try push memory high, you can try playing with PWM Frequency and memory OCP tuning. Also there are three empty footprints for tantalum caps, so I usually solder pair of 330uF 4V caps there.

PLL voltage modification

Simple LDO with feedback adjustment , each one powers separate curcuitry in GPU, but at default they set to provide very close voltage at about 1.05V. So no special magic here.

Cold modification

As was with majority of NVIDIA cards since G80 there is always some thermal shutdown curcuitry present to disable power when
temperature reaches overheating threshold level. It's good thing for safety to prevent GPU burnout if you forgot to put your fan, but
not so good for subzero overclocking. So to prevent power shutdown when temperature freezes below -70c need to disable overtemperature protection.

This can be done by removing tiny resistor shown on photo above. If you want to revert it back - you can just put a short there, that resistor is 0 ohm anyway on reference cards.

Power limiter modification

To fool power limitation set by NVIDIA for kepler's we can reduce resistance of current shunts. Reference card uses three 5 mOhm resistors to measure input power, if we short them or solder on extra current shunt in parralel we can decrease power draw values reported to GPU. This is a must if you want to avoid capping by power and TDP.

So that's it, not very complicated even for beginners. Usual practice to check GPU, memory and PLL resistance before powering up card, just to make sure nothing shorted by accident. Good resistance for GPU is in range 1.5-10 ohm, for memory 100-140 ohm, for PLL hundred or so ohm.

For those who tired of removing shims on other GPU solutions, there is good news for GTX 680 - no need to remove anything,
protective metal shim surface is thinner than GPU die, so just any KPC TEK-9 LN2 containers will fit perfectly on GPUs using
standard mounting.

Crazy LN2 overclockers part

Toggling VID each time and risking to get OVP/OCP with stock VRM is not always a best option, so if someone really want to push
card to limits and beyound, then using external power for card is a choice. GTX 680 draws so little power comparted to overclocked GTX 580, that using wires to hook extra power board have some practical sense. I was measured ~270-300W power draw when card running 3Dmark11 P at 1400MHz with 1.4V, what is really good result for it's performance.

And power plane for GPU voltage is placed on outer surfaces of PCB , what is making attachment of EPOWER really easy job for 30 min of time.

I think pics of actual process with say better than thousand of words.

To be able to run card with EPOWER only one modification for stock VRM needed - cut one trace marked below red.

Make sure that you connect EPOWER by shortest wires possible, and as much wires as possible.

I was little bit busy with a lot of other things, so i did not go crazy with wiring, just bunch of AWG10 grounds and direct PCB-PCB connection for power.

Me and k|ngp|n tested about 6 GTX 680 cards, four of which we tried with EPOWER and got ~1600-1700MHz without problems with only 1.5-1.6V set by EVBOT on EPOWER with only -50 to -60C container temperature. We pushed to max tonight and got 1900 MHz with some little voltage push and -150..-160C. To be honest - that was one of best OC sessions ever with well deserved single card WR in 3Dmark11 P.

We would not stop at this and will push for moar in next days.
Now time to have some rest and go sleep after week of 2-3 hrs sleep.

One more record was broken today as well, we got more than 400 users viewing EVGA's GTX 680 OC experience which can be considered as a good start for future records.

Stay tuned.

Last edited by TiN; 04-05-2012 at 11:13 AM.
Reply With Quote