Sajin
1. Yes.
2. Correct.
3. Yep.
4. Yeah.

 
sweet! thanks a lot for the help :) appreciated!
 
cheers

No problem.

While the guide is great and I have now a few profiles with really nice curves, does anyone have more technical information regarding how the OC Curve works?
 
Like, I note that when you initially setup a curve, the points/dots will spread a few voltage steps on the same clock close to your "target" one. For example, if I have my max (1093) at 2190 clock, the 4 or 5 dots before it will be at 2175.
 
But during tests I noted this would sometimes result on the next step down (2160) to be using low voltage because of this, so in some cases I edited all my last clocks to have like only 2 or 3 voltage steps.
 
But even that doesn't always works. Does anyone know how the interaction o voltage steps <-> clock should work?
 
Also, something not entirely clear to me yet: how boost (in this case 4.0 as I'm on series 20) is suppose to interact with curve OC?
In the majority of my curves I setup them while on idle (35C in my case) so my max/target clock is actually a little higher to what I want, as temperatures will rise and reduce the clock. For example my most stable curve so far is targeted at 2205, which will translate on a real clock of 2175Mhz during benchmarks/games.
 
But ONE of my curves (that I tweaked WHILE benchmark, so temps were at 55) is saved at 2175Mhz on max clock, but when I apply it while idle, it will overclock to 2220Mhz.
 
Is this curve bugged? How is the interaction suppose to be?
 
Thanks

fleptronic
While the guide is great and I have now a few profiles with really nice curves, does anyone have more technical information regarding how the OC Curve works?
 
Like, I note that when you initially setup a curve, the points/dots will spread a few voltage steps on the same clock close to your "target" one. For example, if I have my max (1093) at 2190 clock, the 4 or 5 dots before it will be at 2175.
 
But during tests I noted this would sometimes result on the next step down (2160) to be using low voltage because of this, so in some cases I edited all my last clocks to have like only 2 or 3 voltage steps.
 
But even that doesn't always works. Does anyone know how the interaction o voltage steps <-> clock should work?
 
Also, something not entirely clear to me yet: how boost (in this case 4.0 as I'm on series 20) is suppose to interact with curve OC?
In the majority of my curves I setup them while on idle (35C in my case) so my max/target clock is actually a little higher to what I want, as temperatures will rise and reduce the clock. For example my most stable curve so far is targeted at 2205, which will translate on a real clock of 2175Mhz during benchmarks/games.
 
But ONE of my curves (that I tweaked WHILE benchmark, so temps were at 55) is saved at 2175Mhz on max clock, but when I apply it while idle, it will overclock to 2220Mhz.
 
Is this curve bugged? How is the interaction suppose to be?
 
Thanks

You are hitting voltage and thermal walls. 1093 is not a magic number. My 2080ti's sweet spot is 1072. I stress my card with Adia64 and run video only to find the voltage sweet spot. I also stress the card when setting the voltage curve. This will help with keeping things locked during benching. Also use a good app to monitor voltages and thermals. I like SIV64 works great and is very accurate. Good luck. 

CptSpig
 
You are hitting voltage and thermal walls. 1093 is not a magic number. My 2080ti's sweet spot is 1072. I stress my card with Adia64 and run video only to find the voltage sweet spot. I also stress the card when setting the voltage curve. This will help with keeping things locked during benching. Also use a good app to monitor voltages and thermals. I like SIV64 works great and is very accurate. Good luck. 

Not sure you understood my questions and I found some stuff you said confusing.
 
- We will always hit "voltage wall" using this method, that's why is called "how to force max voltage".
- 1093 is the max allowed voltage on Turing chip, so they all should accept it provided you have a stable chip and proper cooling 
- Boost 4.0 downlock steps are not the same as thermal wall, if you were referring to it. Thermal wall is 88C on this chip.
 
Of course on air is better to lock the voltage a bit lower if the stock cooler can't handle 1093mV and is thermal throttling badly, stable clock is better than have lots of fluctuations even if peaks higher.
 
But thermal wall is not my case. Yes boost 4.0 will downclock a few steps and I'm compensating it with curve tweaks, but 55C on benchmarks is not even close to thermal wall.
 
My curiosity is more about understanding how the voltage curve is suppose to interact with Boost algorithms and how to proper tweak the voltage <->clock multiple dots properly to get the most of it, as so far I have mixed curve behaviors as explained above.
 
Like, why the curve sets same clock on multiple voltages? Is that how SHOULD be? Can tweak it give better results? Should be only ONE clock dot on the voltage you want? 2? 3?
 
This kind of stuff =)
 

fleptronic

CptSpig
 
You are hitting voltage and thermal walls. 1093 is not a magic number. My 2080ti's sweet spot is 1072. I stress my card with Adia64 and run video only to find the voltage sweet spot. I also stress the card when setting the voltage curve. This will help with keeping things locked during benching. Also use a good app to monitor voltages and thermals. I like SIV64 works great and is very accurate. Good luck. 

Not sure you understood my questions and I found some stuff you said confusing.
 
- We will always hit "voltage wall" using this method, that's why is called "how to force max voltage".
- 1093 is the max allowed voltage on Turing chip, so they all should accept it provided you have a stable chip and proper cooling 
- Boost 4.0 downlock steps are not the same as thermal wall, if you were referring to it. Thermal wall is 88C on this chip.
 
Of course on air is better to lock the voltage a bit lower if the stock cooler can't handle 1093mV and is thermal throttling badly, stable clock is better than have lots of fluctuations even if peaks higher.
 
But thermal wall is not my case. Yes boost 4.0 will downclock a few steps and I'm compensating it with curve tweaks, but 55C on benchmarks is not even close to thermal wall.
 
My curiosity is more about understanding how the voltage curve is suppose to interact with Boost algorithms and how to proper tweak the voltage <->clock properly to get the most of it, as so far I have mixed curve behaviors as explained above.
 

I understand everything you are asking using wall was probably the wrong term. The card will step down in thermal increments like 20c, 40c and 60c. So if you have the card pegged at 1093 and your card likes 1072 it will step down sooner. remember voltage creates heat and heat will cause the card to step down. Try to stress the card while you set the voltage curve and it should give you a sold lock.

CptSpig

fleptronic

CptSpig
 
You are hitting voltage and thermal walls. 1093 is not a magic number. My 2080ti's sweet spot is 1072. I stress my card with Adia64 and run video only to find the voltage sweet spot. I also stress the card when setting the voltage curve. This will help with keeping things locked during benching. Also use a good app to monitor voltages and thermals. I like SIV64 works great and is very accurate. Good luck. 

Not sure you understood my questions and I found some stuff you said confusing.
 
- We will always hit "voltage wall" using this method, that's why is called "how to force max voltage".
- 1093 is the max allowed voltage on Turing chip, so they all should accept it provided you have a stable chip and proper cooling 
- Boost 4.0 downlock steps are not the same as thermal wall, if you were referring to it. Thermal wall is 88C on this chip.
 
Of course on air is better to lock the voltage a bit lower if the stock cooler can't handle 1093mV and is thermal throttling badly, stable clock is better than have lots of fluctuations even if peaks higher.
 
But thermal wall is not my case. Yes boost 4.0 will downclock a few steps and I'm compensating it with curve tweaks, but 55C on benchmarks is not even close to thermal wall.
 
My curiosity is more about understanding how the voltage curve is suppose to interact with Boost algorithms and how to proper tweak the voltage <->clock properly to get the most of it, as so far I have mixed curve behaviors as explained above.
 

I understand everything you are asking using wall was probably the wrong term. The card will step down in thermal increments like 20c, 40c and 60c. So if you have the card pegged at 1093 and your card likes 1072 it will step down sooner. remember voltage creates heat and heat will cause the card to step down. Try to stress the card while you set the voltage curve and it should give you a sold lock.

So you are saying is best to setup the OC curve while stressing the card / max temps (which is what I did on that specific profile I mentioned)?
Interesting, will see how it goes and what results I get.
 

fleptronic

CptSpig

fleptronic

CptSpig
 
You are hitting voltage and thermal walls. 1093 is not a magic number. My 2080ti's sweet spot is 1072. I stress my card with Adia64 and run video only to find the voltage sweet spot. I also stress the card when setting the voltage curve. This will help with keeping things locked during benching. Also use a good app to monitor voltages and thermals. I like SIV64 works great and is very accurate. Good luck. 

Not sure you understood my questions and I found some stuff you said confusing.
 
- We will always hit "voltage wall" using this method, that's why is called "how to force max voltage".
- 1093 is the max allowed voltage on Turing chip, so they all should accept it provided you have a stable chip and proper cooling 
- Boost 4.0 downlock steps are not the same as thermal wall, if you were referring to it. Thermal wall is 88C on this chip.
 
Of course on air is better to lock the voltage a bit lower if the stock cooler can't handle 1093mV and is thermal throttling badly, stable clock is better than have lots of fluctuations even if peaks higher.
 
But thermal wall is not my case. Yes boost 4.0 will downclock a few steps and I'm compensating it with curve tweaks, but 55C on benchmarks is not even close to thermal wall.
 
My curiosity is more about understanding how the voltage curve is suppose to interact with Boost algorithms and how to proper tweak the voltage <->clock properly to get the most of it, as so far I have mixed curve behaviors as explained above.
 

I understand everything you are asking using wall was probably the wrong term. The card will step down in thermal increments like 20c, 40c and 60c. So if you have the card pegged at 1093 and your card likes 1072 it will step down sooner. remember voltage creates heat and heat will cause the card to step down. Try to stress the card while you set the voltage curve and it should give you a sold lock.

So you are saying is best to setup the OC curve while stressing the card / max temps (which is what I did on that specific profile I mentioned)?
Interesting, will see how it goes and what results I get.
 

I had the same problem as you are having. If you stress the card while monitoring your curve you can see it boost. I like Adia64 because it's windowed so you can see afterburner and siv64 all at the same time. Now you can see what the voltage and temps are doing while stressing the card. With that info you can see where the curve peeks and flattens out to establish your sweet spot. When you just use 1093 some of the cards don't like that voltage lock and the dots will decrease trying to lower your lock when you are stressing the card like benching. I also hit apply after each dot I set to -1 from the natural curve. Once you get this dialed in it works great!

CptSpig
 
I had the same problem as you are having. If you stress the card while monitoring your curve you can see it boost. I like Adia64 because it's windowed so you can see afterburner and siv64 all at the same time. Now you can see what the voltage and temps are doing while stressing the card. With that info you can see where the curve peeks and flattens out to establish your sweet spot. When you just use 1093 some of the cards don't like that voltage lock and the dots will decrease trying to lower your lock when you are stressing the card like benching. I also hit apply after each dot I set to -1 from the natural curve. Once you get this dialed in it works great!

I don't use Aida as it's GPU Stress test doesn't really heat up my card, it doesn't even get to 50C and TDP never goes above 80.
So I used Heaven Benchmark instead, as the free version allows windowed mode and keeps running even focusing MSI AB/Curve editor above it.
 
I tested some curve tweaks while stressing the card and it was interesting to discover some stuff (at least on my card)
 
- setting 1093mV will always trigger VRel on perfCap.
- sometimes VRel will also be 100% uptime depending on which clock the next dot after your max-target is. Ex: at some point I was lock at 2160Mhz on 1081mV @ 55C and VRel was 100% uptime. And no matter what, even dragging the curve down to 2130 VRel was still up, only going really low it stopped. It didn't made any sense to me so I went back and had the idea to changed the 1087mv dot to 2175, and just like that perfCap went idle. Not sure why and I didn't managed to get this work 100% of the time tough.
 
So far I was able to do 2160Mhz on 1068mV at 55C, and even bumped the memory a bit more to to 8275Mhz.
 
But there's a catch doing this way: my card idles at 35C, so when I apply this curve before starting a benchmark, it will boost to 2190 at 1068mV, which is too much of course and some benchmarks will crash right away.
 
I wish we could just disable boost because right now is just not being helpful at all.
 
I'll keep doing more tests.

Well I think I'm done with testings and tweaks, ended up with 3 profiles.


1) 
Core clock: 2160Mhz while under 54C and 2145Mhz on benchmarks once hits 55C.
Memory: 8250MhzThis is the one Time Spy will run, so I consider the one to use with "sensitive" engines.
It's pretty solid, will dip below 2145Mhz on TS only for some seconds on the really heavy parts where TS max TDP.

2) 
Core clock: 2175Mhz while under 54C, 2160Mhz if above.
Memory: 8250Mhz
So far tested The Division 2 with it and didn't crashed after a long session, 50C max temp so 2175Mhz stable.
I consider this the "default" one as this game has a sensitive engine.
Time Spy do not run on this thought, will crash mid run.
 
3) 
Core clock: 2190Mhz under 54C, 2175Mhz above.
Memory: 8286Mhz
Superposition 1.1 will run on this, but I suspect sensitive games wont sustain long runs.
Probably the one at the verge of stability as I see some occasional flashes on superposition run, but no crash.
 
All profiles at 1087mV.
50C for gaming max, 56 for benchs
No perfCap, only occasional Power on TimeSpy run.