How to Overclock the GTX 1070, Pascal GP104, & GTX 1080
Overclocking has changed with Pascal. It's gotten more advanced – but not more complex. The core functions of overclocking remain the same; we're still using manual stepping and sliders to achieve high, sustained overclocks, but a new auto OC tool (currently only with EVGA Precision) makes limit finding easier. In theory, anyway. We couldn't get it to work with our GTX 1080, but it seems that EVGA Precision's auto-OC (called “ScanOC”) is now functioning with the latest press drivers and installation package.
But we decided to keep going the manual route – it's just better for testing.
From our GTX 1080 review:
“GTX 1080 overclocking is revised heavily from previous generations. The GTX 1080 and GP104 move nVidia from GPU Boost 2.0 to GPU Boost 3.0. GPU Boost 3.0 and supporting applications from AIB partners will allow for frequency-voltage curve creation that fine-tunes voltages against various frequencies. Rather than functioning on a single slider for frequency, applications like Precision and Afterburner will now create a curve that fits your card's particular silicon.
Anyone who's overclocked knows that a high frequency may be sustainable in most games or for long stretches of time, but that some specific applications or use cases will crash the drivers resultant of the overclock. The obvious solution is to step-down the OC globally until something is stable across the board, but then we're losing OC room in the use cases that didn't originally trigger faults.
GPU Boost 3.0 supports scanners, which figure out the optimal frequency for each individual voltage point. If a particular voltage point is less stable, the frequency offset can be adjusted by the tool or by manual control. In EVGA Precision, a ScanOC tool finds the optimal setup and provides a baseline that allows enthusiasts to further tune their setup. Vendors can determine parameters for voltage-frequency curve determination by the OC scanner.”
GTX 1070 Overclocking Results
Here are the stepped results for our manual overclocking efforts with the GTX 1070:
| core clock | core offset | mem clock | mem offset | voltage | voltage offset | power target | 5m test | Endurance |
| 1771 | 4006 | 950mv | 0% | 100% | P | - | ||
| 1936 | 150 | 4151 | 300 | 1000mv | 30% | 112% (max) | P | - |
| 1974 | 200 | 4353 | 350 | 1000 | 30 | 112 | P | - |
| x | 250 | x | 400 | 1000 | 30 | 112 | F | - |
| 1974 | 200 | 4374 | 375 | 1000 | 30 | 112 | P | - |
| x | 250 | x | 375 | x | 30 | 112 | F | - |
| 2015 | 225 | 4374 | 375 | 1000 | 30 | 112 | F | - |
| 1974 | 200 | 4374 | 375 | 1000 | 30 | 112 | P | - |
| 1974 | 200 | 4404 | 400 | 1000 | 30 | 112 | P | - |
| 1974 | 200 | 4455 | 450 | 1000 | 30 | 112 | P | - |
| 1987 | 215 | 4498 | 500 | 1000 | 30 | 112 | P | - |
| 1987 | 215 | 4551 | 550 | 1000 | 30 | 112 | P | - |
| 1987 | 220 | 4551 | 550 | 1000 | 30 | 112 | F | - |
| 1987 | 215 | 4608 | 600 | 1000 | 30 | 112 | P | P |
| 1987 | 215 | 4704 | 700 | 1000 | 30 | 112 | artifacts | - |
We actually had trouble forcing the GTX 1070 to fail with our initial overclocks. Our ultimate clock-rate landed us at 1987MHz core, 4608MHz memory (GDDR5). Voltage never exceeded 1000mV, even if we went beyond a 30% voltage offset (VBIOS restriction for safety, as usual with all cards). Power percent target capped at 112% on our GPU. The final results operated under a 215MHz core offset and 600MHz memory offset.
GTX 1080 Overclocking Results
For reference, here are our GTX 1080 overclocking results:
| Core Clock (MHz) | Core Offset (MHz) | Mem CLK (MHz) | Mem Offset (MHz) | Power Target (%) | 5m Test | 60m Burn-in |
| 1733 | 5005 | 100 | Pass | Pass | ||
| 1880 | 150 | 5305 | 300 | 120 | Pass | - |
| 1974 | 171 | 5305 | 300 | 120 | Pass | - |
| 1987 | 190 | 5305 | 300 | 120 | Pass | - |
| 2025 | 210 | 5305 | 300 | 120 | Pass | - |
| 2050 | 230 | 5305 | 300 | 120 | Fail Driver Crash |
|
| ~2030 | 220 | 5400 | 400 | 120 | Pass | Pass |
GTX 1080 Overclocking Results with GN DIY Hybrid Solution
| Core Clock (MHz) | Core Offset (MHz) | Mem CLK (MHz) | Mem Offset (MHz) | Power Target (%) | 5m Test | 60m Burn-in |
| ~2050-2072 | 225 | 5005 | 120 | Pass | - | |
| ~2100-2138 | 250 | 5005 | 120 | Pass | - | |
| 2164 | 275 | 5005 | 120 | Pass | - | |
| 2164 | 275 | 5602 | 600 | 120 | Pass | Pass |
| 2189 | 300 | 5602 | 600 | 120 | Fail | - |
| 2189 | 300 | 4489 | -500 | 120 | Pass | Pass (But struggled – so would probably fail with longer burn-in; failed immediately when gaming / not synthetic load) |
| 2202 | 325 | 4489 | -500 | 120 | Pass | Fail |
| 2164 | 275 | 5602 | 600 | 120 | Pass | Pass |
GTX 1070 Overclocked FPS vs. Founders Edition, GTX 1080, etc.
And a look at FPS impact:
At 4K in GTA V, our GTX 1070 experiences a 12.6% performance hike in average FPS (GTX 1080 FE saw a 14.88% gain). Mordor at 4K benefits from a 29.9% performance gain – the game is fairly clock-intensive, and the GTX 1070 doesn't seem to be hitting the memory bottlenecks that we found on the 1080. At 1440p, Mordor gains ~23.3% over the stock clock. DOOM at 1080p shows a move into 120Hz range with our overclock, pushing 125.7 FPS against the 114.7FPS stock (+9.2%). At 1440p, that gain is 21.8%, with noteworthy gains in 1% and 0.1% low framerates.
Some games extracting an additional >20% from an overclock – especially with the stock cooler and PCB – is a tremendous boost to performance and a positive indication for the 1070's future. AIB partners will undoubtedly ship pre-overclocked versions, and user overclocking headroom is tremendous – and with real gains. We don't seem to be hitting some of the same concerns as with the GTX 1080, likely due to the slightly cooler nature of the chip.
Continue to the last page for the conclusion.
