GTX285 various mods.. which may work for GTX260/280/295 also

[LexDiamonds]

Interesting mod, but video cards are not 12v car amplifiers... it would seem the initial load on the FETs from charging the caps would eventually kill the card. It is one thing to reduce ripple, but another to overload the card with capacitance....

[NapalmV5]

well last night after true120 copper install (stupid idea) everything went to hell

at first the card didnt boot.. after adjusting the cooler i got it to boot.. the weight of the cooler pulled on the backplate after i removed it i noticed one of the small ceramic cap on the back of the gpu might have been touching the metal plate.. or at least it looked like it im not certain.. most im concern about.. as for the caps ive recapped several times to no available.. the card is in bad shape.. everything gained was lost even default clocks stability is gone lol kaos!

hope the ceramic cap @ the back of the gpu is fine.. recapping to tantalum caps is what i wouldve done in the end anyways.. the next mod was gonna be the inductors but im gonna leave it that for now

let you guys know in a couple of days how the recapping to tantalum caps goes

btw.. have been playing around with a zotac amp 285.. vcore modded @ same voltage as ssc.. its doing exactly the same: 750/1700/1490 only 10mhz less on memory just a slight v increase to match ssc's voltage/clocks is all it needs

[dengyong]

well last night after true120 copper install (stupid idea) everything went to hell

at first the card didnt boot.. after adjusting the cooler i got it to boot.. the weight of the cooler pulled on the backplate after i removed it i noticed one of the small ceramic cap on the back of the gpu might have been touching the metal plate.. or at least it looked like it im not certain.. most im concern about.. as for the caps ive recapped several times to no available.. the card is in bad shape.. everything gained was lost even default clocks stability is gone lol kaos!

hope the ceramic cap @ the back of the gpu is fine.. recapping to tantalum caps is what i wouldve done in the end anyways.. the next mod was gonna be the inductors but im gonna leave it that for now

let you guys know in a couple of days how the recapping to tantalum caps goes

btw.. have been playing around with a zotac amp 285.. vcore modded @ same voltage as ssc.. its doing exactly the same: 750/1700/1490 only 10mhz less on memory just a slight v increase to match ssc's voltage/clocks is all it needs

Must be luck of the draw cause my vanilla zotac does over 1500mem stable with no mods. (yet)

[NapalmV5]

^ nice.. what core/shader @ ?

yeh im not surprised.. its luck of the draw since all 285s use the same memory

[dengyong]

^ nice.. what core/shader @ ?

yeh im not surprised.. its luck of the draw since all 285s use the same memory

Using precision I run 753/1675/1507

I think the memory will do a bit more but I haven't tested.

[NapalmV5]

i precision set @ those clocks and when i run ETQW the system reboots lol vcore modded!!

looks like i got the vanilla and you got the amp for a vice versa $$$ not fair!!

[dengyong]

i precision set @ those clocks and when i run ETQW the system reboots lol vcore modded!!

looks like i got the vanilla and you got the amp for a vice versa $$$ not fair!!

I'm getting a full cover block for it as soon as they're available, I have to turn the fan up to 90% to play crysis for more than 15min.

[NapalmV5]

water should help get 50/50/50 + on top of mods


amp @ 1.26v i can game ETQW stable @ 800/1764/1485 @ precision.. no cap mods

though a few mem mhz less thats smgt my ssc is not capable of @ same voltage without the cap mods

[NapalmV5]

deleted

[NapalmV5]

update: ssc gpu is fine.. ssc is back!

i just tried 4 x 820uf 4V chemicon caps where 4 x 330uf tantalum caps normally occupy and now at least stock clocks are stable again

now i can breathe easier.. whephew

time to find out what really makes the 285 tick

[Biker]

He he he....

I think that poor SSC needs to be booked in for some trauma support

[kiwi]

Napalm, have you tried RT softmod to overcome OVP?

I like your true120 copper on GPU, almost 2Kg is pretty extreme

[NapalmV5]

^ lol its finally out of intensive trauma unit and doing well now

^ unfortunately isl6327 + rt: no go


clocks back @ ~800/1750/1500 @ ETQW

no loss @ clocks/performance

as for gain.. instead of the bulky caps on the back.. these kemet caps perform just as good and keep temps down

the cornell fim caps are bulky themselves and the leads would be even longer so i didnt even bother.. the small film caps would go better

[NapalmV5]

vmem


10K : 2.25-2.29v
4.7K : 2.34-2.37v

[CryptiK]

Those caps look much better, although having as much value as the card in caps added to it was good for the extreme factor If they give so much benefit (30MHz core) I wonder why for the sake of saving a few cents Nvidia didn't put caps on the vacant spaces in the first place.

Just to clarify, was it the addition of the caps on the 4 vacant spots on the back of the card that gave the initial 30MHz increase in core? Then the swapping of the solid caps for the 1000uf 16v blackgate fk's that gave the extra 20MHz?

Nice work on the vmem mod too

[NapalmV5]

thanks

yeh the caps on the back and front pushed the ssc further.. it wouldnt budge no matter what.. vcore only .03v higher didnt help anything.. neither the cooler

without the caps 800/1750/1500 no way in hell @ ETQW

[ptelles]

Nice work Napalm!

[NapalmV5]

thanks


well guys i believe this is the ovp mofo:



pretty easy to find thanks to the isl6327 datasheet

the 100K resistor rings all the bells to me.. what do you guys think?

it would have to be removed and replaced with a ~50k or less or vr

can you guys confirm? if someone confirms ill replace the damn thing in a heartbeat

[CryptiK]

I know you know what you are doing, but I have to ask are you certain it is OVP triggering the shut down though? That is, if you raise vGPU during idle, does the card get a black screen and vGPU ~0v or reboot when ~1.30v+ is reached? Or is it only under load it results in black screen or shutdown? If its the second option it is most likely OCP being triggered, unless under load there is .

I'd need to see a close up pic of the area surrounding the ISL627 chip, but from the circuit diagram it appears OVP is controlled by one pin (44), and you may very well be correct with your identification. Can you locate pin 44 and follow where the trace from that pin leads?

To completely remove OVP it says to leave the pin open so removing the resistor completely would remove OVP. However if you wanted to increase OVP point you could add a 100K resistor in parallel giving 50K total, giving a 50% reduction in OVP signal, or you could use 100K VR in parallel and tune total resistance to allow and adjustable OVP point.

Doing the OCP mod is harder obviously, as you'd need to do something similar to what I did with the G80 OCP/vdroop mod which was solder SMC resistor across each capacitor in each phases R/C circuit controlling Isen for each phase (linked to Isen+ & - pins for each phase).

[NapalmV5]

pin 44 doesnt lead to anything on the front so it leads to the back.. right to that resistor.. at least how i see it


even @ 10k/15k resistor the vcore still remains @ 1.26 and the card doesnt os boot.. i say it has to be ovp kicking in.. i dont think ocp would be active before the ovp at this voltage

thanks for the help.. ill try that ^

[CryptiK]

Ok sure, can you measure if there is continuity between pin 44 and one end of that 100K resistor? (I'm not sure if the pins are large enough to probe individually). If so, I'd go for it.

If it won't even load OS @ 1.26v then I'd agree its probably OVP triggering shutdown. It would IMO be safer to add a 500K VR to the 100K resistor in parallel so you'd only be getting around 83K total resistance reducing the OVP signal by ~17%, so if its not the correct resistor it wont be causing a very large alteration in circuit behaviour. If this at least lets you load OS at 1.26v then you know you're on the right track, so reduce it to 50K and try again.

[NapalmV5]

i added a 100k resistor and now the cores finally getting 1.3v @ 10k but no os boot

added 47k 22k 10k (1.32vcore) no os boot

ill try 200k next

edit: @ 200k added to the 100k vcore @ 1.27 @ 10k

still no os boot

[CryptiK]

Do you mean you'll now remove the existing 100K resistor and put in its place a 200K resistor?

The minimum OVP point with a valid VID is VID + 150mv, so if EVGA have set VID to 1.20v OVP should not trigger until 1.35v minimum. However if they are adjusting vgpu above the set vid by bios or another method, while retaining the standard VID of 1.15v, that puts OVP at 1.30v like is happening.

Since that lowering that resistors resistance is having the effect of raising the vgpu the core receives, but even adding 10K to it decreasing OVP signal by ~91% didn't allow OS boot at only 1.32v, perhaps it's worth considering removing it entirely leaving the OVP pin open? This isn't 'safe' but shouldn't cause any negative effects as pin 44 is only used for OVP. If that doesn't allow boot at any voltage, something else must be limiting it.

That's if adding the 200K resistor doesn't work.

[NapalmV5]

yeh i was thinking the same.. removal

im doing this on the amp 285

havent tried it on the ssc yet

[CryptiK]

It doesn't show on the circuit diagram what OVP pin connects to. The pins that are connected to ground through a resistor is pin 11 (I out), TCOMP, OFS, FS, SS. SS is pin 43, right next to OVP.

Sure pin 44 is connected to that 100K resistor?

The only real way to work this out is find continuity between pin 44 and a nearby component with the DMM.