[SOLVED] How to properly OC an RX 580 Sapphire Nitro+ ?

Feb 7, 2018
79
1
545
2
I have read a guide online on how to OC, and for the most part it sounds right but some parts I do not understand. I am fairly new to GPU OC'ing so go easy on me. I posted a picture of my MSI Afterburner base, and then my MSI Afterburner with the OC that they told me to put. this is the website btw https://www.online-tech-tips.com/computer-tips/overclock-gpu-safely-boost-performance/ . I've done what they said but I dont see that big of a performance increase or temperature increase really. Please help me out, thanks.

Pics of MSI (First is normal, second is OC)

View: https://imgur.com/a/1JrLPQX


View: https://imgur.com/a/yHTug40
 
1237mV is far too high for 1411MHz. It should be stable around 1150mV, maybe a tad lower lower, but probably not less than 1100mV. YMMV, silicon lottery at play.

Are you using anything (HWiNFO64 or GPUz) to monitor your clockspeeds? I'd be willing to bet you're hitting thermal and/or power limits under load @ 1237mV.
Various people have different programs they like to use to test OC stability. My personal favorite is (believe it or not) is Folding@Home.

Here's a guide I put together (RX580 is just an OC'd RX480)
 
Last edited:
1237mV is far too high for 1411MHz. It should be stable around 1150mV, maybe a tad lower lower, but probably not less than 1100mV. YMMV, silicon lottery at play.

Are you using anything (HWiNFO64 or GPUz) to monitor your clockspeeds? I'd be willing to bet you're hitting thermal and/or power limits under load @ 1237mV.
Various people have different programs they like to use to test OC stability. My personal favorite is (believe it or not) is Folding@Home.

Here's a guide I put together (RX580 is just an OC'd RX480)
 
Last edited:
Feb 7, 2018
79
1
545
2
1237mV is far too high for 1411MHz. It should be stable around 1150mV, maybe a tad lower lower, but probably not less than 1100mV. YMMV, silicon lottery at play.

Are you using anything (HWiNFO64 or GPUz) to monitor your clockspeeds? I'd be willing to bet you're hitting thermal and/or power limits under load @ 1237mV.
Various people have different programs they like to use to test OC stability. My personal favorite is (believe it or not) is Folding@Home.

Here's a guide I put together (RX580 is just an OC'd RX480)
No i'm not using anything to monitor my speeds. To be honest I just wanted better performance so I looked up a tutorial on how to OC a gpu and the link I followed was the one mentioned up top. I have no idea what the 1237 mV is or does, as I am particularly new to the OC'ing scene.
 
Feb 7, 2018
79
1
545
2
Read the guide I linked. That will explain a lot. Then feel free to ask ?s.
Alright, I just read the first post on the link you sent. I think Im understanding what undervolting is. So basically you are trying to cut the mV to a lower number so that it is stable but also still getting the same results? and mV means memory voltage right?
 
That's correct, "Auto" voltage generally applies 50-75mV too much at a given frequency on AMD Polaris cards. More voltage = more power/heat/noise.
When I OC (CPU or GPU) I like to use "auto" voltage first. It's helpful to find an initial frequency limit without having to worry about insufficient voltage (shoot the howitzer first). Then dial down voltage at that frequency to find the minimum. Then bump up frequency and voltage by the same %. If unstable, add voltage. If stable, try subtracting a bit of voltage to get back to your "curve". Rinse repeat.

mV = millivolts
 
Feb 7, 2018
79
1
545
2
That's correct, "Auto" voltage generally applies 50-75mV too much at a given frequency on AMD Polaris cards. More voltage = more power/heat/noise.
When I OC (CPU or GPU) I like to use "auto" voltage first. It's helpful to find an initial frequency limit without having to worry about insufficient voltage (shoot the howitzer first). Then dial down voltage at that frequency to find the minimum. Then bump up frequency and voltage by the same %. If unstable, add voltage. If stable, try subtracting a bit of voltage to get back to your "curve". Rinse repeat.

mV = millivolts
Alright I’m about to write a paragraph to make sure I’m doing exactly what you’re saying but how will I test if the volt is unstable? Will it automatically show or would I run a benchmark or something? And to test say stage one I have to set stage 1-7 as the same frequency and volts?
 
Last edited:
I wouldn't bother testing states 1-3. They're all going to be below the voltage limit set by the VRAM, so graph out your results from 4, 6, and 7 then extrapolate the rest.

I use Folding@Home (GPU client only) to test my voltage settings. I've found that it detects problems faster & better than most. 3DMark Time Spy is another decent free program to test with.

I use GPUz to monitor frequency/ temps/ voltage. HWiNFO64 is another alternative.
 
Feb 7, 2018
79
1
545
2
I wouldn't bother testing states 1-3. They're all going to be below the voltage limit set by the VRAM, so graph out your results from 4, 6, and 7 then extrapolate the rest.

I use Folding@Home (GPU client only) to test my voltage settings. I've found that it detects problems faster & better than most. 3DMark Time Spy is another decent free program to test with.

I use GPUz to monitor frequency/ temps/ voltage. HWiNFO64 is another alternative.
Alright, So first off I will test Stage 4 and set 4-7 same freq and volt. I will turn the volt down by about 30 each run. I will test this by running FAHBench after setting the volts. If nothing becomes unstable, I will go down another 20-30 volts, test again if anything happens (unstable) then I will revert the volts back to what they were. Then I will add a percentage such as .25 percent to both numbers, if unstable I will add more volts, if stable Ill reduce volts. And I will do that for each stage. This will in turn OC my gpu. So is there anything I got wrong lol?
 
Sounds right. Overclocking is a bunch of "guess" and check.

If you would prefer an easier solution. List out the frequency of each performance state here and I'll give you something to start with. Then you only have to test smaller changes to get your initial curve before OCing.

From a quick google search, it would seem you can expect to get up to 1500MHz. Not many results are talking about clocks much higher than that.
 
Last edited:
Feb 7, 2018
79
1
545
2
Sounds right. Overclocking is a bunch of "guess" and check.

If you would prefer an easier solution. List out the frequency of each performance state here and I'll give you something to start with. Then you only have to test smaller changes to get your initial curve before OCing.

From a quick google search, it would seem you can expect to get up to 1500MHz. Not many results are talking about clocks much higher than that.
Here's the settings

View: https://imgur.com/a/rKo1HWW
 
600MHz/820mV
900MHz/865mV
1145MHz/920mV (test memory voltage with states 4-7 set at these values)
1215MHz/940mV
1255MHz/960mV
1300MHz/1010mV
1400MHz/1120mV

Keep memory frequency at 2000MHz and set memory voltage to 930mV and Memory Timing to "Level 2". If it doesn't work, try 940mV.

I haven't tested much above 1400MHz, but at the trajectory of my data, you should be somewhere in the 1230mV range for 1500MHz.
 

ASK THE COMMUNITY

TRENDING THREADS