Easy method to predict OC capabilities in current Nvidia GPU's

[SF3D]

Nice chart Sascha!

But we need to remember, that there is other variables which will affect overclockability as well.
Biggest effect will come from temperature, so I usually test in same air temperature in my work room. If room gets few degrees hotter, it can cause some difference.

In the end, the chart looks exactly like expected. Thanks again!

[Mauro]

0.925 - 870 - NA - NA - Asus 470 - Mauro

Tried Heaven again with extreme tesselation and everything maxed out on 1680*1050.
Vent was on 100%, voltage 1,087V and memory at stock.

Passed successfully with core on 880 without any artifacts, when core was set at 885 test failed.
So, at Heaven maxed out stable would be core on 880.

[PiLsY]

Not getting great results under water from the PNY card. Memory is poor - maxes at 1925 and I cant get the core over 895 without furmark hardlocking. Could be down to running sli and not synching the clock speeds between the cards.

I saw a large gain going to water form the other 2 cards - I think this is down to the huge drop in heat. The third card ran some 20c cooler on air though so Im leaning towards the smaller drop in temp giving me less of an oc increase.

[Systemlord]

My GTX480 (P0 level 3 as set in Inspector) seems its at it stable limit at 870MHz @1.138 with memory OC to 4200MHz, but if I try 880MHz with same vcore stressing with EVGA OC Scanner I hear electrical zapping sounds coming from the card, doesn't happen with OCCT though. It isn't capacitor squeal either! I understand that cards with a low PO value OC better, but then why are we able to change the P0 level in Inspector?

-Systemlord

[cowie]

Another 460 hawk i tested,this one goes with SF3D's theory.
P0 1.037

AIR 893 1.21V LN2 1080 1.27v

Highest P0 volt card and the lamest one of of three i tested air and cold.

[QuadDamage]

Review card. lidless. P0=0.962V.

970/1940/2200 at 1.275V, plain water, GPU block only.

http://img215.imageshack.us/f/30632.jpg/

[SF3D]

I had some time today, so I put my cards in to identical situation and tested the max clockspeeds with watercooling (very simple loop and small radiator)
The interesting thing is, that when you push your card to the maximum clock and you do not use fixed voltage, cards start to act more unpredictable. (quite obvious)
Well, that is only partially true. There is thermal shutdown mechanism still active on these cards, even all other protection is shutted off with modifications and software.
The cards which can scale with voltage are in the top of the chart and the cards which have thermal issues are on the bottom. (see reference / lightning)



Now it is easy to pick up 4 cards to my next 4-way SLI overclocking session

[rintamarotta]

Hi all!

I have been testing several GTX 480 reference cards on aircooling and LN2.
I usually always test cards on aircooling and then on LN2.
This time I found something new on Nvidia cards.

On current generation cards you can easily see the P0 performance level voltage.
Just open the Nvidia inspector and open the overclocking page.
On the slider on right you can see P0 performance level voltage.
If it does not work you can check it with multimeter in 3D load etc.



P0 is very interesting value,
cause it will tell us few important things.
Lower voltage means directly lower leakage,
cause signal can be delivered inside GPU with less energy.
Higher P0 voltage will mean that the core need more voltage for same data output,
so more energy is lost and more unnecessary heat generated.

GPU's with low P0 value will work cooler and what I have been told from my sources,
there is higher thermal shutdown treshold on these GPU's as well.
So, if you need to pick some GPU for testing,
try to find low P0 ones.

This is nothing that new for most of us,
but now it is easy to see and predict how your card will overclock.
There is easy way to compare GPU's and it is fixed voltage.
If you use for example 1.07V voltage and 100% fan speed max overclock is correlating with P0 value in most cases.





Now I would need to ask you guys put some input on this thread:

Reference GTX 480 designs use 1.07V and 100% fan speed. Then try to run 3Dmark vantage and report the OC results here.

What I have seen in my own tests:
P0=1.063V / MAX OC on air = 775MHz in 3DMark Vantage
P0=0.985V / MAX OC on air = 900MHz in 3DMark Vantage

P0=1.063V / MAX OC on LN2 = 1100MHz in 3DMark Vantage
P0=0.985V / MAX OC on LN2 = 1300MHz in 3DMark Vantage

This thread is here to help you pick up the best cores for air, h2o and LN2 overclocking.
I hope at least few will bother to run few tests and let us all know the data.
This is confirmed information already from my sources, so there is no need to try to argue

P.S This prediction method is proven to work on GTX 460/470/480 cards. It should work same way in all GPU's which have P0 3D VID values + powerstates.

How about 0.937V P0? 1000Mhz MaxOC? I got that one in my GTX465 (later i flashed it with gtx470 bios successfully).
Only thing that limits the oc atm is my PSU on my HTPC, might need hook this up to my intel machine later for more testing.

At the moment 850/1700/2000 but my 500W Antec PSU on HTPC cannot hold much more load on 12volt rails, those rails are allready at nearly its OCP limit at 34A @ 12v.

[Systemlord]

I had some time today, so I put my cards in to identical situation and tested the max clockspeeds with watercooling (very simple loop and small radiator)
The interesting thing is, that when you push your card to the maximum clock and you do not use fixed voltage, cards start to act more unpredictable. (quite obvious)
Well, that is only partially true. There is thermal shutdown mechanism still active on these cards, even all other protection is shutted off with modifications and software.
The cards which can scale with voltage are in the top of the chart and the cards which have thermal issues are on the bottom. (see reference / lightning)



Now it is easy to pick up 4 cards to my next 4-way SLI overclocking session

My GTX480 doesn't even come close to the voltage you used (P=0 1.063), I'm at 850Mhz, 1700MHz and 4400MHz at 1.138V. When at 880MHz my card starts buzzing and whisling to a point where I don't care to hear that crap, my temps approach 60C at 880MHz!

-Systemlord

[Gaul]

Sparkle GTX 470, P0 = 0.975

need v1.025 on 800 mhz to pass furmark

[rintamarotta]

GTX465 flashed to GTX470 tested with my main pc's psu, P0 at 0.937 and 940mhz @ 1.087 and 4200 on memory dont have time to start flash again to switch volt limits further, quite possible that this chip can reach 1000mhz with something like 1.113v or 1.125 or then i hit brick wall and scaling goes bad.

[Gaul]

[SF3D]

My GTX480 doesn't even come close to the voltage you used (P=0 1.063), I'm at 850Mhz, 1700MHz and 4400MHz at 1.138V

If you use VID 1.138 it is even more than I used on my tests. Check the table again. You need to read voltage in real time during 3D load. P0 1.063V and 1.138VID should give around 1.4V in 3D load

No wonder the pwm is making some sound there

[QuadDamage]

Yeah the juice is usually 0.05V higher on my card than what i set in AB and above 1.3V it becomes super unpredictable. After my last Vantage benching session my card had deteriorted to the point it couldn't even do a measly 800Mhz (monitor shut off) but after a couple hours it started clocking again. Weird.

SF3D, one question, these cards scale real nice with extra clocks in Vantage but beyond 940Mhz core i wasn't getting any better GT1 scores, any clue as to what was holding me back? Even at 970Mhz i wasn't seeing any gains.

[SF3D]

Yeah the juice is usually 0.05V higher on my card than what i set in AB and above 1.3V it becomes super unpredictable. After my last Vantage benching session my card had deteriorted to the point it couldn't even do a measly 800Mhz (monitor shut off) but after a couple hours it started clocking again. Weird.

SF3D, one question, these cards scale real nice with extra clocks in Vantage but beyond 940Mhz core i wasn't getting any better GT1 scores, any clue as to what was holding me back? Even at 970Mhz i wasn't seeing any gains.

Remember, that you need to measure voltage from pcb during load, to be able to tell what it is for real. If you set 1.275V in afterburner, it is something near 1.4V based on your P0 VID. These cards really can not handle 1.4V+ on aircooling because cores are drawing too much current. When you cool it down to -150C current draw is a lot lower, so the PWM can handle more V.

Your card is near 1GHz, so it is coming really unstable or you are hitting minor CPU bottleneck. I guess it is the instability of the vga, cause CPU is making only small difference, when you are 1200MHz+.

[Systemlord]

If you use VID 1.138 it is even more than I used on my tests. Check the table again. You need to read voltage in real time during 3D load. P0 1.063V and 1.138VID should give around 1.4V in 3D load

No wonder the pwm is making some sound there

3DMark06 Score 15,138 @ 850MHz, 1700MHz Shader and 4400MHz Memory.

I don't understand what it is you are saying, checking the table does exactly what? When I reset MSI Afterburner my Core Voltage (mV) is preset at 1.063V. My current overclock is unstable any lower than 1.138V. If I run EVGA OC Scanner or 3DMark06 these seem to trigger the buzzing sounds, esspecially if I forget to click "Unlock Power Draw" sometimes but not always. Is there something going on with my card? I think you need to treat me as a beginner, I'm new at overclocking! What the heck is VID and PWM? Huh?

-Systemlord

[QuadDamage]

Remember, that you need to measure voltage from pcb during load, to be able to tell what it is for real. If you set 1.275V in afterburner, it is something near 1.4V based on your P0 VID. These cards really can not handle 1.4V+ on aircooling because cores are drawing too much current. When you cool it down to -150C current draw is a lot lower, so the PWM can handle more V.

Your card is near 1GHz, so it is coming really unstable or you are hitting minor CPU bottleneck. I guess it is the instability of the vga, cause CPU is making only small difference, when you are 1200MHz+.

Yeah i know man, i always use the MM. And actually 1.07V in afterburner is 1.12V real in 3D and drops to 1.01V in desktop. I'm on water by the way. Been atleast 5 years since i last fired up an VPC unit for subzero 3D benching goodness. Miss the old days.

[Timoleros]

My results for my Sparke GTX480 are:

On stock cooling
P0=0.987V
914/1828/2000 @ 1.13V passed 3DMark Vantage (maybe go higher with OCP mod)
930/1860/1980 @ 1.13V passed 3DMark 06

I will try with w/c these days...

[fordf250]

Is 1.075V safe on air? The evga gtx470 I have has .937V po and does 750mhz with no voltage change but only 825mhz with a bump to .987V. I was hoping it would do over 900mhz when I read this thread. Testing now at 1.075V and 900mhz.

[Systemlord]

Is 1.075V safe on air? The evga gtx470 I have has .937V po and does 750mhz with no voltage change but only 825mhz with a bump to .987V. I was hoping it would do over 900mhz when I read this thread. Testing now at 1.075V and 900mhz.

On air cooling with good case air flow using Afterburner you can max out at the highest voltage Afterburn will go as long as you are within thermal limits, since the GTX 470 overclocks cooler, better and easier than the 480 I don't see an issue and long as you note you highest temperature after you overclock you card.

-Systemlord

[saaya]

wow that card5 looks weird petri!
looks like its sucking loads of current, so thats how it can clock as well as others with lower volts, but ocp kicks in sooner...
i wonder how that gpu would do with a better pwm

hmmm so maybe comparing current draw can tell you something about how well a card will probably clock on ln2 than checking how high it clocks at x volts?

[Systemlord]

If you use VID 1.138 it is even more than I used on my tests. Check the table again. You need to read voltage in real time during 3D load. P0 1.063V and 1.138VID should give around 1.4V in 3D load

No wonder the pwm is making some sound there

Purhaps you can help me here, the 480 card that I RMA'ed had P0=1.063v, my new replacement card is even better at P0=0.988v, but buzzes and squeals as if it were the same exact card! I strongly believe something else is causing my cards to buzz and squeal, both cards which are clearly different cards are acting like cloans in when and how they buzz and squeal at exactly the same MHz and while leaving them at stock voltages.

Systemlord

[Nanok]

What do you guys think about this wounder ?

[Systemlord]

What do you guys think about this wounder ?

I'm not sure what wonder you are referring to, you haven't even overclocked it yet! Your idle voltage is nothing special at a P12 state, now it it was at .962v with P-State at P0 then you would have something special. If your stock voltage is 1.113v then that is the worst VID I have ever seen!

-Systemlord

[VoodooProphetII]

Okay my Inno3d GTX 480 stock voltage is 0.988V. At stock I can push it up to 800 and run the standard 3dmark benchmarks. At 1.075V I can go up to 880mhz core on stock air at 100% fan. At 1.1V I can run all of the 3d marks at 900mhz core and the highest the temps has gone with the stock cooler was 69C with the Accelero XTREME cooler it only went as high as 59C with 1.1Vgpu and the core at 900mhz while running benchmarks.