Overclock.net › Forums › Volt Mods › General Volt Mods › Graphics Cards Volt Mods › GTX 670 reference PCB - voltmod
New Posts  All Forums:Forum Nav:

GTX 670 reference PCB - voltmod - Page 4

post #31 of 94
Thread Starter 
Glad to tell you guys that I just did the VFB mod at home.. and it works =)
The OVP limit is anywhere between 1.3625-1.4125V (1.395V typical according to datasheet, while I use this unlocked bios), but that is easily enough for me. I'm considering 1.3V max voltage, if the core clocks well to that point. If the core hits the wall somewhere in the progress, I'll stop and be happy with what I got... But I'm hoping for 1400MHz, let's see =)

I soldered a wires between VFB and GND leg of an output capacitor. In the end of wire is a 2-pin female connector with 2.54mm raster.
I had accidentally dropped the VR I took from work.. so I had no VR to test with. However I had some 1206 chip resistors, which I then soldered on pin header.. This header is easy to swap. =)

No VFB-GND resistor gave me 1.236V, which is a bit more than what the VID's have set. This could also be error from my 7€ multimeter biggrin.gif I'm going to borrow a more accurate one from work for the weekend..
150k resistor gave 1.244V
100k --> 1.249V
75K --> 1.254V
50K --> 1.266V

I ran heaven benchmark once with 1.266V. I set the core 1372MHz and memory was 3758MHz.. It worked like a charm.. I'm going to try higher clocks now..
Gonna take another VR tomorrow and do some more testing. =)

Edit: Just ran through with 1385MHz. It's not fully stable though. First time didn't go through. I should try without clocking the memory next, because I think I'm pushing them quite far here. 3758MHz sounds like a lot tongue.gif I'm not sure what they usually are.

Somethings wrong with EVGA Precision X though! It won't let me increase core clock by +1. It jumped from 1372 to 1385.. Tried every value in between, but it stayed 1372 for many increments and then jumped to 1385.. Should I use another program?
Edited by Olper - 11/28/12 at 11:39am
post #32 of 94

..maybe your hitting the next strap level with core OC.. a lot of GPU's jump in small increments at a time. So long as you see your specs increase in GPU-Z when you make the change.. you should be fine [i dont think its a glitch].

 

 Glad to see you got things working! Nice clocks there too! Check your voltage droop for +12v input, GPUv and vMEM.. may be able to squeeze some more with a bit more capacitance in the right spots. thumb.gif

post #33 of 94
Looking good thumb.gif


I didn't quite understand your question about the VID. If you want to work it out, probe the connections for the VID assignments and what resistors they are connected to, then mark them in an image.
    
CPUMotherboardGraphicsRAM
E8400, E8600, 980BE T2RS+, P5E3, TA890FXE 8800GTS's,GTX's,GS,Ultra's,9600GT Lots of D9's and some crap DDR3 
Hard DrivePowerCase
Seagate 250GB + 80GB WD for benching TT 875w, Corsair 650TX Custom bench station 
  hide details  
Reply
    
CPUMotherboardGraphicsRAM
E8400, E8600, 980BE T2RS+, P5E3, TA890FXE 8800GTS's,GTX's,GS,Ultra's,9600GT Lots of D9's and some crap DDR3 
Hard DrivePowerCase
Seagate 250GB + 80GB WD for benching TT 875w, Corsair 650TX Custom bench station 
  hide details  
Reply
post #34 of 94
Thread Starter 
I'll leave the VIDs alone.. I don't need them. Also it's wise not to play with them because bios controls them too. You have to cut the trace or remove components to do that..


I had a wild idea about making a PCB with tiny pogo pins. The pogo pins would touch VFB, GND and GPUv. Then on the PCB would be a microcontroller to control a digital potentiometer. The controller could read the GPUv and adjust the potentiometer... Then I could connect the microcontroller by USB inside the computer-case.. wouldn't have to solder anything on a 400€ pcb and wouldn't have to open the case to adjust voltage =)

It's just an idea for now..
post #35 of 94
Thread Starter 
Upped the voltage to 1.3V. I can get 1411MHz steady without overclocking memory. When I overclock memory to 3758MHz it starts throttling because of power%. Power target is 112%, but it throttles every time I hit over 105%... don't know why..

Maybe I'll cheat the shunt resistor.. but first I must find the right one tongue.gif I bet it's the one on front side. Near the PCI-E connector
http://www.techpowerup.com/reviews/Palit/GeForce_GTX_670_Jet_Stream/images/front_full.jpg

Edit: If I command voltage to 1.2125V at EVGA precision X, the voltage rises to 1.27V. When the card is under load, voltage rises to 1.30V... Overshooting droop compensation?
Edited by Olper - 11/29/12 at 11:37am
post #36 of 94

See if these have continuity to each other.. I may have the order swapped.. but you get the idea

 

 

1000

 

 

 

**and some good spots to land more polymer caps too !! Looks like 2x easy spots for GPUv and 2x for vMEM

post #37 of 94
Thread Starter 
According to TiN's guide, on 680 board there is three current shunts. I can find three on 670 also. Two of them are on the front, which means I'd have to remove the waterblock frown.gif
Let's see if I get to it this weekend..

By the way, I ran 3dmark11 with 1398MHz. The score was worse than with 1340MHz.. I wasn't logging, but I'm sure there was power% throttling.. will confirm this later today
post #38 of 94
Thread Starter 
Confirmed that 3dmark11 throttles a lot. It goes almost down to 1200mhz during tests.. https://www.dropbox.com/s/fpv0k4ofky1h9uv/throttling.png

By the way, do you guys think this kind of droop is normal? It raises the voltage by 30mV.. Maybe it would be better that there was more voltage with less droop compensation. It's best this way with most CPUs I think..
Edited by Olper - 11/30/12 at 11:24am
post #39 of 94

Ive had identical PCB/core/rev GPU's.. 1x would droop .03v on load.. the other would raise +.06v on load..  I dont know if there is anything to it other than the core or perhaps BIOS? That particular phenomenon has always been attributed to 'randomness' in my book though. biggrin.gif Cant say that I was ever able to find a source for the difference from card to card..

post #40 of 94
Thread Starter 
Got some 2512 size 4mohm resistors here. The originals are 5mohm.. if I replace these, I get good power% information. 100% --> 80% etc.

Three shunt resistors make sense. One per each 6pin connector and one for PCI-E connector. Maybe I should do something to the overcurrent protection while I'm at it.

I ordered some voltmeters for monitoring tongue.gif
Don't know how accurate these are by default, but calibrating should be easy enough..
http://www.aliexpress.com/snapshot/208357283.html
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Graphics Cards Volt Mods
Overclock.net › Forums › Volt Mods › General Volt Mods › Graphics Cards Volt Mods › GTX 670 reference PCB - voltmod