GTX 1080 GPU Boost 3.0 fail?

Joined
Dec 21, 2016
Messages
11
Okay so before anyone hates me, hear me out. So I recently got a new GPU, the GTX 1080 Waterforce edition from Gigabyte. And of course the first thing I wanted to know, is how far I can clock it, since it's watercooled. I was rather quickly dissapointed. But nonetheless I kept trying to tweak my way to a higher clock, and I tried playing with the GPU boost curve tool. And that is when I found something rather odd. When I used the curve tool to just overclock the last voltage step, 1.093 that is, and set the clock to 2164MHz. I came out with a great result, I got it to run stable on that clock at low temps (55 degrees under load) every game and no artifacts. So I tried to run 2177MHz on that voltage and I'm still quite happy with the results. I managed to get it running 90% stable which on such a low voltage, most can't even run the GPU with that clock. Anyway back to the main point, I then tried overclocking the GPU using the slider in Afterburner (Tried other programs too) And set it to +110MHz which at 1.093 volts puts me at 2164MHz. Here's the problem, every game I would play I would artifact or crash or both. I tried to lower the slider to +100 and that set me at 2152MHz and although it was 99% stable and didn't crash,it still, rarely would artifact. Just on a side note if I overclock the last voltage step to 2164MHz and just even slightly increase the clock on any other voltage step (Doesn't matter which one, even the first one you can adjust) then it does this. I don't get why but I'm wondering if anyone can answer as to why it happens or if anyone else can confirm this for me (As in try and see if you can achive a higher and more stable clock than you already do, the same way I did, not try to get the same clocks as me)
 
Last edited:
Yeap, welcome to GPU boost 3.0 - where you can use the old method and mostly get mediocre results, or actually put some work in with the new method and get better results.

I'm in the same boat with my 1080 ACX 3.0 (base model) that I put under an AIO cooler. 2050 MHz is around the max stable frequency using the old method. My max stable OC with the new method? 2202MHz! Nuts, right?
 
Yeap, welcome to GPU boost 3.0 - where you can use the old method and mostly get mediocre results, or actually put some work in with the new method and get better results.

I'm in the same boat with my 1080 ACX 3.0 (base model) that I put under an AIO cooler. 2050 MHz is around the max stable frequency using the old method. My max stable OC with the new method? 2202MHz! Nuts, right?

Isn't Nvidia like aware of this? Shouldn't someone talk to them about this? Because this seems rather stupid to implement something that sounds nice but actually fails to work correctly, taking away performance for an average overclocker. And what about my side note? Like if you think about it that is absurd for a feature to behave that way, especially when we pay so much, to get GPU's like these.
 
Isn't Nvidia like aware of this? Shouldn't someone talk to them about this? Because this seems rather stupid to implement something that sounds nice but actually fails to work correctly, taking away performance for an average overclocker. And what about my side note? Like if you think about it that is absurd for a feature to behave that way, especially when we pay so much, to get GPU's like these.

Huh? It works the old way just fine. If you want BETTER results you use the new way.

What's wrong with that, ya stick in the mud!:ROFLMAO:
 
Huh? It works the old way just fine. If you want BETTER results you use the new way.

What's wrong with that, ya stick in the mud!:ROFLMAO:

What's wrong? Maybe that overclocking using the slider to 2164MHz makes me artifact and crash, while using the curve tool it doesn't? However that works only when I overclock the last voltage step (1.093) on its own, leaving me unable to touch any of the other voltage steps, becasue that would then make me artifact and crash again? I think you misunderstood what I meant. Like is it not weird, that on the same voltage with the same settings and same clock, one way works partially only when it's just that one single voltage step (1.093) being adjusted (Touching others just a slight bit makes me artifact and crash again) and the other method (Slider) doesn't work at all because it just artifacts and crashes? And why does it do that? As in why does it artifact and crash using the curve tool if I adjust the last voltage step, and any other voltage steps with it, even a little bit?
 
Last edited:
Yeap, welcome to GPU boost 3.0 - where you can use the old method and mostly get mediocre results, or actually put some work in with the new method and get better results.

I'm in the same boat with my 1080 ACX 3.0 (base model) that I put under an AIO cooler. 2050 MHz is around the max stable frequency using the old method. My max stable OC with the new method? 2202MHz! Nuts, right?

what's the "new method"?
 
What's wrong? Maybe that overclocking using the slider to 2164MHz makes me artifact and crash, while using the curve tool it doesn't? However that works only when I overclock the last voltage step (1.093) on its own, leaving me unable to touch any of the other voltage steps, becasue that would then make me artifact and crash again? I think you misunderstood what I meant. Like is it not weird, that on the same voltage with the same settings and same clock, one way works partially only when it's just that one single voltage step (1.093) being adjusted (Touching others just a slight bit makes me artifact and crash again) and the other method (Slider) doesn't work at all because it just artifacts and crashes? And why does it do that? As in why does it artifact and crash using the curve tool if I adjust the last voltage step, and any other voltage steps with it, even a little bit?

I guess I just don't see it as a problem. GPU Boost 3.0 is a bit different under the hood, being much more granular in how it handles clockspeeds and voltages even when compared with GPU Boost 2.0. With a new system comes a new way to doing things sometimes - like coming from the old GTX 400 or 500 series to 600+ series and wondering what the fuck GPU boost was and having to learn how to overclock with it.


what's the "new method"?

The new Voltage/Frequency Curve editor in MSI afterburner (or PrecisionX if you are a masochist). Only for GPU Boost 3.0 cards.
 
I guess I just don't see it as a problem. GPU Boost 3.0 is a bit different under the hood, being much more granular in how it handles clockspeeds and voltages even when compared with GPU Boost 2.0. With a new system comes a new way to doing things sometimes - like coming from the old GTX 400 or 500 series to 600+ series and wondering what the fuck GPU boost was and having to learn how to overclock with it.




The new Voltage/Frequency Curve editor in MSI afterburner (or PrecisionX if you are a masochist). Only for GPU Boost 3.0 cards.

See you don't see it as a problem and many others wouldn't either. But What I'm saying is that the system doesn't work the way it should, if it does what it does. Therefore thanks to that I can't overclock all of the GPU steps but only the one, if I want the clock that I managed to get, instead of being able to apply that overclock to the whole gpu (All of the steps). That's my problem, loosing out on an overall overclock of the whole GPU.
 
See you don't see it as a problem and many others wouldn't either. But What I'm saying is that the system doesn't work the way it should, if it does what it does. Therefore thanks to that I can't overclock all of the GPU steps but only the one, if I want the clock that I managed to get, instead of being able to apply that overclock to the whole gpu (All of the steps). That's my problem, loosing out on an overall overclock of the whole GPU.

Do all of the frequency steps have the same voltage?

Not with GPU Boost 3.0 (again, far more granular than previous versions), unless you use the Voltage/Frequency tuner control.

If you lock the voltage not using the Voltage/Frequency tuner somehow, it would work like you want.

Or, you could just use the new tuner and BAM - awesomeness.
 
I actually prefer GPU Boost 3.0. It offers me so much more control, and allows for extremely easy undervolting + overclocking. You can also limit voltage much easier now compared to before, where you had to flash a modified BIOS just to get what you can do now.
 
I'm learning the GPU Boost curve method, but the best I can get stable is +114/+300 which nets me 2100/4450. Any amount of voltage doesn't help getting past that. Card runs cool tho @ 58c under load.
 
Do all of the frequency steps have the same voltage?

Not with GPU Boost 3.0 (again, far more granular than previous versions), unless you use the Voltage/Frequency tuner control.

If you lock the voltage not using the Voltage/Frequency tuner somehow, it would work like you want.

Or, you could just use the new tuner and BAM - awesomeness.

See I still don't think you get what I mean I'll have to show you using photos :D
 
Okay so before anyone hates me, hear me out. So I recently got a new GPU, the GTX 1080 Waterforce edition from Gigabyte. And of course the first thing I wanted to know, is how far I can clock it, since it's watercooled. I was rather quickly dissapointed. But nonetheless I kept trying to tweak my way to a higher clock, and I tried playing with the GPU boost curve tool. And that is when I found something rather odd. When I used the curve tool to just overclock the last voltage step, 1.093 that is, and set the clock to 2164MHz. I came out with a great result, I got it to run stable on that clock at low temps (55 degrees under load) every game and no artifacts. So I tried to run 2177MHz on that voltage and I'm still quite happy with the results. I managed to get it running 90% stable which on such a low voltage, most can't even run the GPU with that clock. Anyway back to the main point, I then tried overclocking the GPU using the slider in Afterburner (Tried other programs too) And set it to +110MHz which at 1.093 volts puts me at 2164MHz. Here's the problem, every game I would play I would artifact or crash or both. I tried to lower the slider to +100 and that set me at 2152MHz and although it was 99% stable and didn't crash,it still, rarely would artifact. Just on a side note if I overclock the last voltage step to 2164MHz and just even slightly increase the clock on any other voltage step (Doesn't matter which one, even the first one you can adjust) then it does this. I don't get why but I'm wondering if anyone can answer as to why it happens or if anyone else can confirm this for me (As in try and see if you can achive a higher and more stable clock than you already do, the same way I did, not try to get the same clocks as me)

That basically means the voltage steps before the last step is unstable, which can happen, depending on how the default frequency curve looks.
 
Screenshot_1.png Screenshot_2.png Screenshot_3.png View attachment 13354 View attachment 13355 View attachment 13353 View attachment 13353
That basically means the voltage steps before the last step is unstable, which can happen, depending on how the default frequency curve looks.

Right I know but here's the thing. If I overclock them all they just artifact right. But if I just overclock one of the steps, doesn't matter which one, but just one on its own to the same clock it was at before it doesn't artifact. Like I can't overclock all the steps if I want a high clock, I can only overclock one step, and it will work perfectly no artifact and be stable. Do you at least get what I mean? Like here's an example I overclock the last step to 2164MHz works fine. If I overclock the last step to 2164MHz and any other step regardless what step to any value even like 20-40MHz then my last step 2164MHz starts to artifact because of the overclock of the other step/s. And that just doesn't make sense. 1st picture is showing what I can do the other 2 showing exactly what I can't do because it artifacts. The 3rd picture shows what I'm talking about here mainly, my last step is 2164MHz and doesn't work, because the previous step has been overclocked to 2152MHz which also on its own works perfectly fine. And it's not the previous steps being unstable because I can run every voltages step at the clock of +110MHz but only if I do it just on one step not all of them.
 
Last edited:
View attachment 13356 View attachment 13357 View attachment 13358 View attachment 13354 View attachment 13355 View attachment 13353 View attachment 13353

Right I know but here's the thing. If I overclock them all they just artifact right. But if I just overclock one of the steps, doesn't matter which one, but just one on its own to the same clock it was at before it doesn't artifact. Like I can't overclock all the steps if I want a high clock, I can only overclock one step, and it will work perfectly no artifact and be stable. Do you at least get what I mean? Like here's an example I overclock the last step to 2164MHz works fine. If I overclock the last step to 2164MHz and any other step regardless what step to any value even like 20-40MHz then my last step 2164MHz starts to artifact because of the overclock of the other step/s. And that just doesn't make sense. 1st picture is showing what I can do the other 2 showing exactly what I can't do because it artifacts. The 3rd picture shows what I'm talking about here mainly, my last step is 2164MHz and doesn't work, because the previous step has been overclocked to 2152MHz which also on its own works perfectly fine. And it's not the previous steps being unstable because I can run every voltages step at the clock of +110MHz but only if I do it just on one step not all of them.

Does the core clock hold? Or does it downclock 1 step (13Mhz)?
 
I adjust the whole curve, not just one dot which is the 1075. Never did it that way so..
 
I adjust the whole curve, not just one dot which is the 1075. Never did it that way so..

Yeah but I can't do that because when I do and I go to 2164MHz then it artifacts, whereas if I adjust just the last step on its own to 2164MHz it works perfectly.
 
I adjust the 1.06V (or so, hard to tell) and higher voltage points, and I leave the lower voltages alone.
 
You guys could try it yourselves. Try just adjusting the last voltage step and see if you can get your cards to a higher stable overclock. Because this to me is really weird and it would be even worse if it was to be the same for everyone else too.
 
Yeah I tried it. Helped me get 2100 on the core and 4500 on the mem currently. It's a little tricky, but the sad part is the voltage is pegged @ 1093mv. Can't seem to get past that :meh:
 
I remember fooling around with this when I first got my 1070. I found that manual overclocking using the curve did allow me to hold a higher core frequency, but for whatever reason my benchmark (Firestrike) scores were consistently lower. I went back to just setting an OC using the slider and haven't bothered with it since.
 
Yeah I tried it. Helped me get 2100 on the core and 4500 on the mem currently. It's a little tricky, but the sad part is the voltage is pegged @ 1093mv. Can't seem to get past that :meh:

So you're confirming that what I said doesn't exsist just for me but for everyone right? Because if you are then that confirms GPU Boost 3.0 is a pile of wank that needs to be fixed. And yeah about the voltage it's the same for all of us until a bios ediotr comes out or a modded bios :/
 
Well it does work, just not the way we want it to work. And yes it does need to be refined and tweaked so it'll be user friendly. It seems most of the 1070's voltage is pegged at more or less the 1093mv mark. Hopefully there will be a fix or a way to edit the bios.
 
Unless you measure with a voltmeter you can't really know what the voltage is - the software voltage reading is very much a ballpark figure.
 
Tried this with my titans and I couldn't get even an extra 5 mhz out of the overclock. I can put +180 on the core and no higher, regardless if the slider or curve is used (even at just the last voltage step)
 
Okay so before anyone hates me, hear me out. So I recently got a new GPU, the GTX 1080 Waterforce edition from Gigabyte. And of course the first thing I wanted to know, is how far I can clock it, since it's watercooled. I was rather quickly dissapointed. But nonetheless I kept trying to tweak my way to a higher clock, and I tried playing with the GPU boost curve tool. And that is when I found something rather odd. When I used the curve tool to just overclock the last voltage step, 1.093 that is, and set the clock to 2164MHz. I came out with a great result, I got it to run stable on that clock at low temps (55 degrees under load) every game and no artifacts. So I tried to run 2177MHz on that voltage and I'm still quite happy with the results. I managed to get it running 90% stable which on such a low voltage, most can't even run the GPU with that clock. Anyway back to the main point, I then tried overclocking the GPU using the slider in Afterburner (Tried other programs too) And set it to +110MHz which at 1.093 volts puts me at 2164MHz. Here's the problem, every game I would play I would artifact or crash or both. I tried to lower the slider to +100 and that set me at 2152MHz and although it was 99% stable and didn't crash,it still, rarely would artifact. Just on a side note if I overclock the last voltage step to 2164MHz and just even slightly increase the clock on any other voltage step (Doesn't matter which one, even the first one you can adjust) then it does this. I don't get why but I'm wondering if anyone can answer as to why it happens or if anyone else can confirm this for me (As in try and see if you can achive a higher and more stable clock than you already do, the same way I did, not try to get the same clocks as me)

Because when you use the slider, boost 3.0 will automatically try to run the clock at the lowest voltage on the curve. If you look at the curve, after you set that offset on the slider, I would imagine the lowest voltage would end up being 1043 or 1050mv. Which is likely too low, especially with the temps you were seeing. By using the voltage / frequency curve to overclock your GPU, you set the lowest voltage at your specific clock to 1093mv. Boost 3.0 didn't have an option to attempt to run at a lower voltage. You see, even using offsets, you're at the mercy of the curve..... Given the temps you were running, frankly, I'm shocked it stayed stable at all.

There are a couple of different methods. If you're looking for 180+, set the curve to 170+, then manually adjust the last 10mhz or so, to the voltage you wish to run at. Like this:

cHLlI5C.png


OR, just grab the 1093mv slider and yank it up to what you want the clock to run at. Like this:

cLQdhSY.png


Both have their pros and cons....but can be equally effective.
 
Last edited:
None of those worked for me. @2200 and 1093mV, I got a hard lock and a DPC_watchdog_Violation :rolleyes:
 
somebody get on the phone with nvidia right now and tell them running their cards above spec isn't working. this is unforgivable.
 
lol, it works just fine. Just have to be patient, and use the voltage / frequency curve for overclocking.
 
Lmao, temps don't exceed 58c under heavy benching, so temps are not an issue. I'm going to call nvidia in the morning and give them a piece of my mind! This is unacceptable!
 
Working with boost 3.0, temps are ALWAYS an issue. Every 10c or so, it will try to either downclock or raise voltage to keep the GPU stable. Those thermal layers start at 15c (of the ones I've found), and WILL cause problems if you're not paying attention to them. This isn't the grand ole days where ya just threw some voltage at it, and some offset and called it a day. You're gonna have to work for it, by managing voltage, core clock and temps. lol

Some pretty amazing clocks can be had by managing those 3 things properly.

geCW1mm.png
 
As stated above, temps are not an issue with my card. Not the first time in this rodeo. I'm well aware of how to clock these newer gen cards. I'm telling you overclocking using the curve does nothing for me. I get better results using the slider method. Using curve best I can get is 2074. Anything past that will result in hard locks. Using sliders I'm able to get 2138 and it'll downclock to 2100-ish.

1093mV is not enough voltage for these cards.
 

Attachments

  • Screenshot (11).png
    Screenshot (11).png
    31.9 KB · Views: 82
A little more voltage, and you might see a little bit more clock, but you're also going to see more drops in core clock because the additional voltage will just add heat. This really isn't hard to understand. You have to manage heat, voltage, and frequency if you want to get a decent overclock. By using the voltage / frequency curve, you can lock in a specific voltage for a specific clock. It won't throttle, but, if temps start to creep up over one of those "thermal layers" that happen every 10c or so, it may start to become unstable, drop frame rates, or just crash all together.

Do what ya wish, but.......you're never going to hit a really good overclock on pascal without using the curve. And.....56c is too warm for a high overclock, btw.

Best o' luck to ya
 
I knew I had a screenshot of it somewhere. This is a 1080 FTW on air, ambient temps at about 18c and 100% fan speed. I used Firestrike Ultra graphics test 1 and 2. The ridiculous curve I used is also pictured. That's 2227 on air. The score wasn't as good as 2153 on air, but it ran it. The higher temps make the core run less efficient. Lower temps, lower voltages required for a specific core clock to run good, and stable.

Just tryin to help

NmaA0La.jpg
 
I hear ya, but the 1080 has GDDRX ram, higher boost clock and more cuda cores. Very few 1070's can reach the same clock speeds. I'm surprised my card can reach 2100 as most of them will top out at 2050 or so. My specific card refuses to clock any higher than 2100 no matter the method. I'm sure with that needed bump in voltage, it will tip the scale. Have you seen the cooler my card has? Even at fans @ 100% I can barely hear it, and temps are well within the throttling area.
 
Back
Top