NVIDIA GeForce GTX 1070 Overclocking Review

NVIDIA GeForce GTX 1070 Overclocking:

Now that I have looked at the first two iterations of NVIDIA's GP104 16nm Pascal architecture, the verdict is in and damn is it impressive. What we have is a performance powerhouse that exceeded most of our expectations. The GTX 1070, in stock trim running the dynamically managed GPU Boost 3.0 clock speeds, has a tremendous amount of delivered FPS performance in each resolution. The stock clock speeds for the GP104 core equipped with 1920 CUDA cores on the GTX 1070 are 1506MHz with a GPU Boost 3.0 clock speed of 1683MHz. The 1683MHz is a pretty decent boost out of the gate, but there is more to be had. Truth be told, in some scenarios you may see more than 1683MHz in game when the thermal and power limits are not exceeded.  Manging clock speeds dynamically, NVIDIA uses GPU Boost 3.0 technology to cycle between the base and GPU Boost 3.0 clock speeds. We, however, can usually grab some free performance up and over those dynamically managed limits.

In the images below, you can see that when clock speed offsets are adjusted on a linear curve, you have the potential to lose performance based on the inability to modify individual voltage frequency offsets to fill those lost gaps. However, out with the old and in with the new. To take advantage of this technology, each of the usual suspects are putting together new enhancements to their overclocking tools to take advantage of this opportunity. Right now we have the latest Alpha version of EVGA's Precision XOC tool. It's not yet ready for prime time, but it does allow the end user to tweak and tune the clock speeds using multiple formats with a built-in OC scanner.

EVGA's Precision XOC tool is the application we will use to complete the overclocking on the GTX 1070. If you have used it before, many of the controls will look very familiar to you. However, with the implementation of GPU Boost 3.0, we get a couple new wrinkles. The control mechanisms are well laid out and are fairly self explanatory. Where we get into some differences are when you move to the tuning section. Here is where you will have the option to set up either a basic offset voltage frequency curve, a linear offset V/F curve, or manually tweak your own voltage frequency curve.

The settings menu is pretty brief, but it does allow you to tune the OC scanner voltage and frequency offests to shrink the overall changes when you get to the final manual tuning stages. When you tune manually, there is a built-in OC Scanner that uses what looks like a FurMark derivative to scan for graphic artifacts. I found that using the built-in OC Scanner tool I ended up pretty close to my ultimate final core clock speeds on this GPU. Further manual tuning ultimately will get you all the performance the card has to offer. By using this tool, the final GPU Boost 3.0 core clock speed was set to 1911MHz. A nice bump over the factory 1683MHz boost speed. However, there is no work on the memory or fan speeds. You will still need to do this work on your own.  

After using the OC Scanner in EVGA's Precision XOC tuning utility to try and find the best voltage/frequency ratio, I figured there would be more clock speed left on the GTX 1070's plate, much like I saw with the GTX 1080. By cranking the fan speed up to 100%, I was able to reach a stable 2050MHz to 2076MHz core clock speed. To do so, you will have to set the power and thermal limit sliders to the maximum, and prioritize the limits to look at temperature first. Exceeding the 91 °C limit won't be an issue with the three heat pipe-based cooling solution, where the power limit seems to be the deciding factor. Keeping the GP104 core cool is paramount to getting the most core clock speed out of it. The 2050MHz speed was the worst case I saw using the OSD in EVGA's tool. Starting benchmarks at 34 °C, the clock speed would scale up to 2126MHz, but start dropping as the thermal load increased with the additional current draw in a heavy 3D load. It's interesting to watch the clock speed drift around in 13MHz increments.

Overall, manual tuning gives you the best shot at the highest core clock speeds. Memory tuning was completed after finding my final core clock speed on the GP104 core. The GTX 1070 is not equipped with GDDR5X and makes use of some high speed Samsung GDDR5 memory. Speeds of 1900MHz to 2000MHz was about all you could get out of the 7Gbps rated GDDR5 memory on the previous generation Maxwell cards. The GDDR5 on this card scaled easily well past that number to 2259MHz or an effective 9000MHz, or a 1000MHz bump over the 8Gbps rating. Pretty solid marks for the second tier card.

Ultimately, I used EVGA's Precision XOC tool to get a solid baseline and took the overclocking from there, since the tool does not force the issue on the GDDR5 memory. Using clock speed offsets of +227MHz on the core and +522MHz on the memory, I was able to achieve a solid 22% bump on the core and a 13% bump to the GDDR5 memory. That's free FPS performance just ripe for the taking.

Testing Setup:

• Processor: Intel Sixth Generation Core i7 6700K @ 4.5GHz
• CPU Cooling: Custom water cooling from Swiftech
• Motherboard: ASUS ROG Maximus VIII Extreme
• Memory: G.SKILL Trident Z 16GB 3400Mhz
• Video Cards: NVIDIA GTX 1070 Founders Edition
• Power Supply: Corsair AX1200
• Hard Drive: Corsair Force GT 240GB SSD x2
• Optical Drive: Lite-On Blu-ray
• Case: Corsair Obsidian 650D
• OS: Windows 7 Professional 64-bit

Comparison Video Cards:

• XFX R9 390X DD
• MSI GTX 970 Gaming 4G
• PNY GTX 980 XLR8
• XFX R9 Fury
• PowerColor R9 390
• NVIDIA GTX 980Ti
• NVIDIA GTX TitanX
• NVIDIA GTX 1080 Founders Edition

Maximum Clock Speeds:

Testing for the maximum clock speed consisted of looping Unigine Heaven 4.0 for thirty minutes each to see where the clock speeds failed when pushed. If the clock speed adjustment failed, then the clock speeds and tests were re-run until they passed a full hour of testing.

• Gaming Tests:

1. Fallout 4
2. Grand Theft Auto V
3. Middle-earth: Shadow of Mordor
4. Far Cry Primal
5. Battlefield 4
6. The Witcher 3: Wild Hunt
7. Hitman (2016)
8. Tom Clancy's The Division
9. Unigine Heaven Benchmark 4.0
10. Ashes of the Singularity
11. 3DMark

• Usage:

1. Temperatures
2. Power Consumption