Gigabyte GTX 1070 G1 Gaming - 24/7 Overclocking Guide (Basic method)

[RodroG]

HOW TO GET A 24/7 OVERCLOCK FOR GIGABYTE GTX 1070 G1 GAMING

***Remember that any overclocking process is risky, so be aware that you try this on your own risk***

A) Setout

What is a 24/7 Overclock (OC)?
In easy words, it is the overclock profile for your graphic card that can be used 24 hours a day, seven days a week and 365 days a year, as it is stable under full load conditions and does not reduce the lifetime of your device.

The process to determine this special overclocking profile is not fast. It requires patience and a lot of time to properly complete all its stages.

How can we trust that we will not reduce the lifetime of our graphic card?
Through two essential preventions:

1. Keep a good GPU temperature, applying a proper fan mode or a custom fan curve that will always keep it between 60 and 70 Celsius degrees.
   
2. Do NOT modify the stock GPU voltage.

How do we measure graphic performance improvement?
Through the following benchmarking and graphic stress programs:

• 3DMark[www.futuremark.com]
  
• Heaven 4.0[unigine.com]
  
• Valley 1.0[unigine.com]
  
• FurMark[www.geeks3d.com]
  
• OCCT[www.ocbase.com]

All of them must be installed and updated before starting the overclocking process.

In addition you can also use these ones:

• EVGA OC Scanner X[www.evga.com] (you must be logged in to download)
  
• MSI Kombustor[www.geeks3d.com] (you must download the 32 bits or 64 bits version depending on your system)

How to get this bench/stress softwares?
It is simple, you can get them by clicking on the names of the list above or searching by their names through your internet browser.

How to check the stability of your OC profile?
Through these same programs using them as graphic stress tests. That is to say:

• 3DMark: Running its Fire Strike Stress Test, consisting of 20 uninterrupted loops of the Fire Strike benchmark Graphic Test 1.
  
• Heaven: Running it in loop mode for at least 30 minutes.
  
• Valley: Running it in loop mode for at least 30 minutes.
  
• FurMark: Running its GPU Stress Test at least 1 hour.
  
• OCCT: Running its GPU stress test at least 30 minutes (tick Error Checking box only when you are testing Memory OC).

Other programs needed
Our main overclocking utility, one GPU monitoring software and one spreadsheet tool.

Specifically, these are the ones I used:

1. Gigabyte - Xtreme Gaming Engine (XGE)[www.gigabyte.com.es]: To configure and apply overclocking profiles.
   
2. TechPowerUp GPU-Z[www.techpowerup.com]: To know key information about our graphic card and its maximum clocks, temperature, TDP and cap or throttling reasons.
   
3. Microsoft Excel[www.microsoftstore.com] or a free equivalent[es.libreoffice.org]).

Instead of the Gigabyte utility you can use other overclocking tools like MSI Afterburner[es.msi.com] or EVGA Precission X[www.evga.com]. Just use the OC program you prefer.

GPU-Z must be open for registering all key data related with the applied OC profile while Heaven, Valley, FurMark and OCCT are running as stress tests.

With Excel we will create a worksheet where we will record all the relevant data for this overclocking process and with which we will identify the most stable OC level. Here[1drv.ms] is a empty but formatted spreadsheet sample you can use.

Process to identify our 24/7 Overclock
The full process consists of 3 main stages:

1. GPU Overclock (GPU OC)
   
2. Memory Overclock (Memory OC)
   
3. Games & 24/7 Overclock (Final)

------------------------------------------------------------------

B) First Stage - GPU Overclock

At this stage we will check the stability / performance of the different GPU OC profiles.

Please, follow the procedure below step by step:

1. Checking the factory profiles stability and its performance scores. This profiles are called Gaming Mode (stock) and OC Mode by Gigabyte. Ensure you through XGE that the default values of these two profiles are applied (by clicking Default or Apply button).
   
2. Set through XGE these common values for each of the following OC variables:
   
   - GPU Voltage: +0%
   - Power Limit: +11% (max)
   - Temperature Limit: -3 ºC (target: 80 ºC)
   - Memory Clock: +0% (stock)
   - FAN: Choose a suitable fan mode or set a custom fan curve. The fan mode or curve we set must keep GPU temperature below 70 Celsius degrees in any stressful graphics conditions.
   
   Record all of these values in the spreadsheet for each OC profile.
   
3. Increase the stock GPU clock by 5% through XGE and then continue with 10 MHz increments per step. Note each increment in our spreadsheet.
   
4. At each OC level start by running the 3DMark Fire Strike Stress Test. This point is passed only if this stress test ends with a result of at least 97% in its Frame Stability Rate (FSR) AND without noticing artifacts during the test.
   Recommendation: during this test look at the screen almost all the time to detect possible artifacts.
   Save the result / report that 3DMark gives you, take note of the maximum temperature shown in the chart and click the Compare Result Online button to get an extra online report. In the online report find the values for FSR, Core Clock and Memory Bus Clock.
   In our worksheet we will write Yes / No in the column called Passed? and all previous data.
   Follow this writing pattern:
   
   - In Passed? write the affirmative results in bold and the negatives in red and bold.
   - Write in bold the clocks that have increased in each level.
   
   
5. If the OC profile seems stable in 3DMark continue by running Heaven for 30 minutes in loop mode with GPU-Z been opened.
   Take a look at the temperature and the maximum theoretical clocks shown by this program during the test to also confirm in this way that the OC profile has been applied.
   Recommendation: Observe the screen repeatedly but only occasionally and for a few minutes to detect possible graphics errors.
   We will close the program if we observe artifacts or if we exceed 30 minutes without seeing errors.
   Whenever we exit to desktop due artifacts or program / driver crashes or when loop time end successfully, we will go to GPU-Z Sensors tab and we will register Max Core, Max Mem Clock and Max Temp values.
   In our worksheet write Yes / No in the column called Stable? and all previous GPU-Z data.
   
6. If it seems stable in Heaven continue by running Valley for 30 minutes and do the same.
   
7. If it seems stable in Valley run FurMark for 1 hour and do the same.
   
8. If it seems stable in FurMark run OCCT for 30 minutes with error checking off and do the same.

Common GPU OC Recommendation: Once you detect instability in any test do not continue testing with the remaining programs.

Common GPU OC Guidelines: Once instability is detected we must record it in the spreadsheet and we will not continue with the other programs, after that we will apply the immediately preceding stable GPU OC profile and we will re-check its stability. If it was stable we will stop and we will stay at this step. If it was not stable we will roll back a step further and we will check its stability again. We will proceed thus until we find a GPU OC level that is stable for sure.

First Stage Final Recommendation: For less risk you should reduce 4-5 MHz extra the maximum stable clock reached and then apply it.

Congratulations! At this point you will have identified the maximum stable GPU OC of your GTX 1070 G1 in stability tests.

NOTE: The 24/7 GPU OC does not necessarily match with the maximum stable GPU OC in stability tests, since it will depend on the third stage of the process that involves intense and long gaming sessions (at least 2 hours) with your most demanding games.

-----------------------------------------------------------------------

C) Second Stage - Memory Overclock

Set our new maximum stable GPU OC, we will check the stability / performance of different Memory OC profiles.

Procedure:

1. Set through XGE these common values for each of the following OC variables:
   
   - GPU Clock: final GPU OC clock identified in Stage 1.
   - GPU Voltage: +0%
   - Power Limit: +11%
   - Temp Limit: -3 ºC
   - FAN: your suitable fan mode or fan custom curve.
   
   Note all this values in your spreadsheet.
   
2. Increase the Memory Clock by 24-48 MHz per step using XGE and then write it down in your Excel spreadsheet.
   
3. Run and observe the Fire Strike Stress Test running and register your data for each Memory OC profile.
   
4. If it was stable, continue with Heaven for 30 minutes with GPU-Z in the background and write all data down.
   
5. If it was stable, continue with Valley ...
   
6. If it was stable, continue in the same way with the rest of programs.
   Common Memory OC Guidelines: Once instability is detected we must record it in the spreadsheet and we will not continue with the other programs, then we will apply the immediately preceding stable Memory OC profile, we will re-check its stability and if it was stable we will stop and we will stay at this step. If it was not stable, we will roll back a step further and we will check its stability again. We proceed thus until we find a Memory OC level that is stable for sure.
   Writing Guidelines: Same as in the previous stage.
   
7. Run 3DMark Fire Strike Benchmark for each stable Memory OC level.
   If the benchmark ends correctly and without seeing artifacts or unexpected crashes, note the overall score and the average frame rate in Graphic Test 1 and 2 and Physics and Combined tests.
   Writing Note: Write in red the overall score when you observe any signs of instability.
   
8. Write in bold the overall scores which are better than those of the first stable Memory OC level and which are also better and at the same time in FPS 1, FPS 2 and FPS Combined partials scores (corresponding to those of the Fire Strike benchmarking tests that are only sensitive to graphic overclocking).
   Note: Sometimes we will check that although we raise the memory clock we get lower scores than a previous stable step.
   Guideline: identify stable Memory OC levels with which the performance improves and between these stay with the one that give you the highest score.
   
9. Final Memory OC setting: Set through XGE the intermediate clock between the best and maximun stable Memory OC profile and the immediately preceding stable profile and test its performance with Fire Strike Benchmark.
   Guidelines: If the benchmark finish successfully note the new profile and its scores, then check if the scores are better than the ones of the preceding best and maximum stable Memory OC profile. If there was an improvement stop the memory overclocking in that new profile and otherwise roll back to the best and maximum stable Memory OC profile.

Congratulations again! You have identified the maximum and best stable Memory OC profile for your GTX 1070 G1 in stability / performance tests.

NOTE: The 24/7 Memory OC does not necessarily match with the maximum and best stable Memory OC in stability / performance tests, since it will depend on the third stage of the process that involves intense and long gaming sessions (at least 2 hours) with your most demanding games.

Personal Note: In my case the maximum and best stable OC profile in stability tests also results stable with Games, so if you have properly followed the steps of this two stages the same thing may happen to you.

--------------------------------------------------------------------------

D) Third Stage - Games & 24/7 Overclock

In this final stage we will check the stability of our maximum and best stable OC profile in stability tests through intense and long gaming sessions with demanding video games.

Setout: Install at least 4-5 demanding video games which together make it possible to check the stability under DX11, DX12, OpenGL and Vulkan API.

Prcedure:

1. Apply through XGE your max stable GPU OC in stability tests and check clock frequencies with GPU-Z.
   
2. Test its stability in each game by 2 hours of uninterrupted play.
   Guidelines: Run the game at least on Ultra and look for artifacts or unexpected crashes. If there are not you can consider the profile stable and you can write it down in your spreadsheet but if there are errors you must reduce -5 MHz the GPU Clock and re-check stability. We will continue this way until we find an stable level.
   For each game and each GPU OC profile add all the OC Settings in your spreadsheet and write down Yes / No in Stable? column.
   The maximum stable GPU OC in Games will be our 24/7 GPU OC.
   
3. With our 24/7 GPU OC been set apply via XGE the max and best stable Memory OC in stability tests and check clocks with GPU-Z.
   
4. Test its stability in each video game by 2 hours of uninterrupted play and note the results.
   Guidelines: Same as previous but if you notice errors lower -12 / -24 MHz the memory clock and re-test stability. We will continue this way until we reach stability.
   For each game and each Memory OC profile add all the OC settings in your spreadsheet and write down Yes / No in Stable? column.
   The maximum stable Memory OC in Games will be our 24/7 Memory OC.

We did it! The union of the maximum stable GPU OC and the maximum stable Memory OC in Games will be the desired 24/7 Overclock for our Gigabyte GeForce GTX 1070 G1 Gaming. Enjoy it!

NOTE: Here is my 24/7 OC profile:

- GPU Clock: +115 MHz
- Memory Clock: +476 MHz
- GPU Voltage: +0%
- Power Limit: +11%
- Temperature Limit: -3 (target: 80 ºC)

Please use it ONLY as a reference. You should find your own 24/7 OC profile! Remember that all Systems and Graphics Cards (also those of the same model) are physically different. These differences make the OC profiles not directly extrapolable. In fact, what was valid for me, does not have to be valid for you.

--------------------------------------------------------------------

I hope you find this guide useful and do not hesitate to comment it!

[[MVM]HunterKiller]

Nice OC mate, I just bought two 1070 G1 Gaming GPUs and I was able to OC with MSI Afterburner

- GPU Clock: +110 MHz
- Memory Clock: +545 MHz
- GPU Voltage: 0 (stock)
- Power Limit: 88% (-12% from 100%)
- Temperature Limit: (target: 78ºC)
- Fan Speed: 60% (fixed for now, sitting at 60ºC)

Both cards are running but one seems to be around 20MHz faster and about ~3ºC cooler.

Now thats running the new 384.94 Nvidia driver W10 64Bit


Have you tinkered with them since?

[RodroG]

Originally posted by MVMHunterKiller:

Nice OC mate, I just bought two 1070 G1 Gaming GPUs and I was able to OC with MSI Afterburner

- GPU Clock: +110 MHz
- Memory Clock: +545 MHz
- GPU Voltage: 0 (stock)
- Power Limit: 88% (-12% from 100%)
- Temperature Limit: (target: 78ºC)
- Fan Speed: 60% (fixed for now, sitting at 60ºC)

Both cards are running but one seems to be around 20MHz faster and about ~3ºC cooler.

Now thats running the new 384.94 Nvidia driver W10 64Bit


Have you tinkered with them since?

Thanks dude! +545 with Afterburner it's a great Mem OC! But why are you lowering the Power limit? Is it because you're ocing two cards instead of one? Usually it is suggested the just the opposite, max. power limit...

I've installed 384.94 but after testing it a bit I roll back to 382.33, I've noticed that the latest driver lowered the idle clock from the usual 240 Mhz to 139 Mhz. Please, could check and tell me back yours with 384.94?

[[MVM]HunterKiller]

Thanks mate, ill definitely look into that, yeah I lowered the power limit because im running 100% load 24/7 so I wanted to get maximum OC for the least amount of power usage.

I also notcied that I can only OC to around +115-125MHz GPU clock memory before it crashes and that doesnt matter on the power limit so I turned the power limit down as low as I could but also maintain the highest clock that I can.

I was also able to OC the memory clock to +645MHz but I didnt check for stability so I just turned it back to where I knew I could get 100% uptime.

To add I was using the driver before this latest one and I still hit maximum head room of about +115-125MHz OC before it crashes.
I'll have to check idle clock later in the week as I won't have time to play around on my PC until then.

[RodroG]

Originally posted by MVMHunterKiller:

Thanks mate, ill definitely look into that, yeah I lowered the power limit because im running 100% load 24/7 so I wanted to get maximum OC for the least amount of power usage.

I also notcied that I can only OC to around +115-125MHz GPU clock memory before it crashes and that doesnt matter on the power limit so I turned the power limit down as low as I could but also maintain the highest clock that I can.

I was also able to OC the memory clock to +645MHz but I didnt check for stability so I just turned it back to where I knew I could get 100% uptime.

To add I was using the driver before this latest one and I still hit maximum head room of about +115-125MHz OC before it crashes.
I'll have to check idle clock later in the week as I won't have time to play around on my PC until then.

Max the power limit it's totally compatible with an OC 24/7. This doesn't affect the lifetime of your card. If you overclock the GPU and the VRAM and you lower the PwLimit less than 100% your card should throttle at load being so difficult to reach and keep the highest and higher clock at boost. Could be an stable OC (not 100% sure IMO) but not with the best performing results. I highly recommend you to follow the guide I did man.

[[MVM]HunterKiller]

I understand what you mean, I'm also trying to reduce power costs too, so that's why I ran the settings I have because it's very close to maximum OC with less power usage overall.
These cards are running 24/7 with OC but if I was running it for gaming then yes I agree to run 100% power plus maximum OC.

[RodroG]

Originally posted by MVMHunterKiller:

I understand what you mean, I'm also trying to reduce power costs too, so that's why I ran the settings I have because it's very close to maximum OC with less power usage overall.
These cards are running 24/7 with OC but if I was running it for gaming then yes I agree to run 100% power plus maximum OC.

First, what I'm saying is a suggestion, so it should be clear you can do what you want with your config . Said that what I'm traying to explain you is that the main point of any 24/7 graphic card overclock is to get the max. and best stable OC in both stabilty tests and playing demanding games. For getting this PwLimit must be maxed (not the Voltage range) and then step by step you should find the extra clock frequencies (GPU and VRAM clocks) which are stable and give you the best performance scores measured through 3DMark FireStrike (DX11) and/or TimeSpy (DX12) graphic benchs (the above OC 24/7 guide explain in detail how to get this via the FireStrike benchmark, for TimeSpy the method is analog).
Currently IMO your OC could be stable but I'm pretty sure it is performing in fact such as the factory stock OC of the 1070 G1 Gaming due to your lowered PwLimit. In other words, your OC clocks are not really working, they are throttled by your current lowered Power level and your cards will perform not according with you OCed clocks values.
Basically you always need an extra Power level or supply for keeping the OC we set. If not, ok, you can see and set the higher/est OC clock values you want but they doesn't work in fact. So a real 24/7 OC must be done and got with the PwLimit maxed and not the opposite. As well and IMO saving power cost would be fine if your current Power supply potential is restricted for keeping an SLI config and you can not keep PwLimit at least at 100% or maxed (here to 111%). A real and effective 24/7 OC should be always Power demanding (not like a punctual extreme OC but anyway it's power demanding), that's the point and should not be a problem for a PC rig which is suitable for SLI configs and GPU overclocking ... Which is your current Power Supply model? A 750 WATT model should be enought for OCing and for setting your SLI config. I hope I have explained better what I wanted to tell you and this can be usefull for you.

[[MVM]HunterKiller]

Ok, ill go into a little more detail, im OCing for mining and not gaming, so im looking at what is the most power efficient for the most stable overclock.

With the power limit being throttled is reducing my overclock is semi true, I am still able to overclock with better hash rates while lowering the power limit, when I set my power limit to 100%, it only makes a very slight difference and not enough to add value to my hashing rate and costing me alot more in electricity.

I was able to get it to
- GPU Clock: +105 MHz
- Memory Clock: +600 MHz
- GPU Voltage: 0 (stock)
- Power Limit: 90%
I could probably go up in GPU clock but I did a PC reboot and I was able to obtain the same clock frequencies when I get the GPU clock to +60MHz, so I dont know if thats because of the Nvidia drivers or something else.

Although the currency im mining at the moment relys heavily on Memory clock and not so much on GPU clock so with this setup it only uses around 86% power MAX from the GPU.

When I mine the other currency it fluctuates the GPU clock and thats where iver throttled the GPU by setting it to 88% because of the reason I stated in my second paragraph.

The GPU's characteristics may and do change with other algorythums so im still playing around with overclocking settings and depending on the algo.

[RodroG]

Ok, if you're doing mining tasks now it makes sense. I'm almost a total ignorant of virtual mining, but from what I've been able to read on this subject and its relationship with OCing the strategy you've followed here makes sense and corresponds to what many mining users are doing.

That said I would like to say that my OCing explanation and recommendation would only be adequate and optimal for getting a 24/7 Overclock for Gaming, never for virtual mining tasks. Thanks for sharing your experience!

[[MVM]HunterKiller]

Yes I totally agree.

[RodroG]

Hi dude! Someone in reddit ask me for the best driver version for mining, have you any idea on this? Thanks in advance

[[MVM]HunterKiller]

To be honest I dont know, I tend to update the drivers as they are released and they all seem to do the same but as I have mentioned above I run at 90% power so im limiting the full OC.

[RodroG]

Originally posted by MVMHunterKiller:

To be honest I dont know, I tend to update the drivers as they are released and they all seem to do the same but as I have mentioned above I run at 90% power so im limiting the full OC.

Ok, thanks anyway.

[kiDk]

HI GUYS,

I am start learning how to overclock my GPU and i saw your guys review, what do your guys recomend for my specs?
I saw [MVM]HunterKiller review i did the settings on GIGABYTE EXTREME ENGINE and i like it, can i do it more? or its ok? more will be risky?

Best regards

MY SPECS:

-AMD Ryzen 7 1800X Octa-Core

-Motherboard MSI B350 GAMING PLUS

-RAM G.SKILL Ripjaws 16GB 3200MHz

-GeForce GTX 1070 G1 Gaming 8GB

-Nox Hummer GD650W

-Philips 273V5LHSB FHD 16:9 60Hz 1ms

-Roccat Kova [+] Gaming 3200DPI Black

-Roccat Ryos MK Pro Gaming Keyboard

-Headsets Hyperx Cloud 2

-Roccat Taito Mid Size 3mm

[[MVM]HunterKiller]

Another good Overclocking program is MSI Afterburner, its the one I prefer to use, I dont think there is an advntage for using it, its just reliable and easy to read.

I have different settings now, compared to the previous post but for gaming, you could push

- GPU Clock: +130 MHz
- Memory Clock: +700 MHz
- GPU Voltage: 0 (stock)
- Power Limit: 100%

Although I have seen clocks higher again
- GPU Clock: +150 MHz
- Memory Clock: +750 MHz
Im just not to sure how stable it is, I would recommend that you start around 120 GPU clock and 650-700 Memory clock then work your way up from there.

For my mining I run the following under 24/7 operation.

- GPU Clock: +120 MHz
- Memory Clock: +650 MHz
- GPU Voltage: 0 (stock)
- Power Limit: 60%
Although i can push just under 130 GPU clock and between 700 and 750 for memory clock but can get a little unstable.

[Monk]

Sorry but on modern cards, you are totally safe to unlock voltage control and max it out along with max temp and power sliders to get the best OC.
You should also start the overclocking process from stock, go up in 10MHz steps for the core and 50MHz steps for the memory, pull back once there's an issue.
The cards cannot pull enough voltage even unlocked to damage themselves, hell myself and other enthusiasts looking g for the ultimate performance even shunt/power mod our cards to get round the very limited voltage controls.

I will say getting an extra 1.3GHz out of your memory is very impressive though.
Though I must say as detailed as your methods are (and everyone has their own system), I feel there is alot of wasted time there.

Also for mining you tend to want the cards undervolted for maximum efficiency not over clocked, unless you don't pay your power bill that is.