Nvidia’s latest Pascal Architecture is one of the hardest pieces of architecture to overclock effectively and reliably. With all the new features and functions Pascal has, they can actually be more of a hindrance than a benefit if all you care about is overclocking.
What Is Overclocking:
Overclocking is when you take a computer component and increase it’s clock speed higher than the rated speed set by the manufacturer. Doing this gives you free performance out of your computer components. The only downside is that you have to stress test o make sure the component is fully functional with the new clock speed. Typically all computer components have some additional amount of “headroom” where you can overclock and the component be fully functional, the reason why is because manufacturer’s always make sure the voltage of a specific component is more than enough for the component to fully operate to make 100% sure that it won’t crash or become unstable. Instability happens when the component is not fed enough voltage to run properly.
GPU Boost 3.0
Lets start with the one feature that makes it so difficult to get good overclocks, and that is Nvidia’s latest GPU boost revision 3.0.
The beauty of GPU boost 3.0 is that right out of the box, it will auto overclock your GPU beyond it’s rated boost clock speed. So as an example, lets say we have a GTX 1080 that has a stock speed of 1500mhz and a boost clock of 1700mhz. Well, GPU boost 3.0 figures out how much power it can use and how much thermal headroom there is, and can core clock up an additional 200mhz. Going all the way to 1900mhz during optimal temps. Now this is just an example, normally GTX 1080’s reach the 1950mhz range, but it all varies per card.
The disadvantage of this kind of feature for overclocking is that there isn’t much headroom left, GPU boost 3.0 already did that for you (which I personally don’t mind, but it can be a bummer for you hardcore enthusiasts). Also, because GPU boost 3.0 varies the frequency based on power limit and temperature, you never get a static peak voltage and clock speed, making stress testing a bit more annoying than it could be.
Now there is a way around this. Fortunately for us, Nvidia allows us to manually manipulate GPU Boost 3.0 to our will. So you can have a static clock speed and static voltage if you want. However be warned, overclocking this way takes WAY WAY longer to do, and usually isn’t worth it.
My recommendation is to leave the manual curve alone and just adjust the core and memory offsets themselves.
How to Overclock:
So before you can do anything, you need a GPU overclocking program to actually do any overclocking. My recommendation is that you use MSI Afterburner 4.3.0 (or later). You can also use EVGA’s Precision XOC but I find MSI Afterburner to be a bit more reliable and simpler. However, if you are using a EVGA graphics card with an ICX cooler, you’ll want to use Precision XOC exclusively to get those extra temperatures on the memory and VRMs.
For this guide, I highly HIGHLY recommend you use Unigine Superposition. It is the most demanding benchmark out right now, and will push your hardware to it’s knees no matter how high performance it is. I also recommend you install Unigine Valley for memory overclocking.
Note: I will not be overvolting in this tutorial. Overvolting on Pascal yeields very little gain in MHz, but will reduce the lifespan of your card by quite a bit if your maxing it out.
So opening up your GPU overclocking utility you’ll see a bunch of sliders:
- Push the Power Limit and Temperature Limit to the max on the slider. This way we won’t get any thermal/power throttling from the card.
- Set a fix fan speed of about 70-100%. Depending on the cooler, set a fixed fan speed so the GPU temps don’t exceed 80C. (Run a simple stress test to see what temps are at stock fan speed, then adjust accordingly.)
- Now you are ready to overclock!
With Pascal, the vast majority of cards can easily hit 50mhz+ on the core from the start. So what you’ll want to do is raise your core offset slider to +50mhz. This will add 50mhz to the core clock.
Next, stress test, if it’s stable for 10 minutes, great! Add another 10mhz and keep repeating. Once you crash, lower the offset by 10mhz and stress test for 2 hours. If it’s stable then great! You have a perfectly reliable overclock (for the most part). Next, you’ll want to play all the games you play on it and make sure it’s stable in those aswell (just play the way you always play, that’s more than enough to stress the GPU).
Overall, the max you’ll probably see is 50-80mhz+ on GTX 1080 Tis, 1080s, and 1070s. With GTX 1060s you’ll see probably get to an extra 70-100mhz, then for GTX 1050/1050 Tis, you’ll see 150mhz+.
This is where the meat and the horsepower lay. VRAM overclocking alone will yield far more FPS than core overclocking on Pascal, but it can also be far more difficult to stabilize vs the core.
The reason why memory overclocking is difficult on Pascal is because of Pascals amazing ability to correct memory errors on the fly. When you push your memory clock so far that it goes unstable, it won’t crash (unless you push it REALLY FAR), instead what you’ll see is a decrease in your FPS. This is because the memory is actually unstable, however Pascal is correcting all the errors the memory is making on the fly which is hogging up resources, which in turn, decreases FPS.
Here is a little trick in figuring out how far you can push your memory clock:
- Open up a game or benchmark that’s relatively new, but at the same time allows your GPU to push beyond 100fps. I recommend Unigine Valley.
- Once you find the 100+FPS sweet spot, go ahead and run it in a windowed mode so you can switch to MSI afterburner/Precision XOC at the same time.
- Watch your FPS counter carefully! See where your average FPS lay.
- Next, increase your memory clock in 100+mhz increments.
- Once you start seeing your FPS decreasing (ever so slightly) or not increasing at all, go ahead and lower it by 100mhz.
- Next stress test for an hour or so. (Just the memory, make sure your GPU has no overclock.)
- If stable, then if you want, you can do this all over again in 25+mhz increments to get the most out of your card, but I’d personally just leave it there. Doing that specific won’t yield more than 1-2 fpss
Adding Core and Memory Overclocks Together:
Typically, simply adding your core and memory overclock should work on the fly. But sometimes it doesn’t. So what you’ll want to do is either game for 2 hours or stress test for 2 hours with both of your max overclocks combined. If stable then great!
If you are getting errors as in textures glitching out but the game is still running, then it’s most likely a memory instability, lower by about 50mhz. If the game crashes, then it’s most likely a core clock instability, decrease by 10mhz.
If it’s still crashing, then decrease both by amounts stated above.
Last but not least, you’ll want to make a fan curve. Doing your own instead of stock allows you to have quite performance when idle, but allows temps to be quite low while gaming. Depending on noise, set your fan curve so that durng max GPU usage the GPU doesn’t pass 72-80C (higher temps allow quieter performance, so it’s your choice).
I hope this guide helped you in overclocking your pascal GPU. On average you should see roughly a 10-15fps increase in gaming performance (depending on the GPU).
GeForce GTX 1080 Overclocking Guide With AfterBurner 4.3.0
GPU Boost 3.0: Finer-Grained Clockspeed Controls
Ultimate How to Overclock Pascal GPU Guide – GTX 1060, GTX 1070, GTX 1080, GTX Titan X