Validating OC stability quickly

[kriegisyourbae]

Hi all,

 

Which tools/settings are best to identify unstable settings in the shortest amount of time possible?

 

Now before you go "google is your friend", please check the following so I can explain myself.

I have so far been using the following tools for stress testing as well as benchmarking:

-Unigine Heaven

-Unigine Valley

-Unigine Superposition

-OCCT

-Furmark

-3D Mark Fire Strike

-3D Mark Time Spy

-Horizon Zero Dawn (Custom settings)

-Metro Exodus (Ultra Preset)

-Rise of the Tomb Raider (Maxed out settings)

 

With none of the tools/game benchmarks used I was able to find reliable results. While all of them were stable at +180 MHz core/+620 MHz memory clock @ 0.912 V, actually playing RotTR resulted in the game crashing. So I lowered the OC slightly and had to repeat this until I was at +160 MHz core boost. The game had crashed up to 3 hours into playing it. Now playing a game for 2-3 hours for each individual value would be insane, that's easily 100 hours with the main purpose of stress testing. So I want to find a method that's more efficient.

 

That's where my question comes in.

 

 Some general information regarding my scenario:

Spoiler

 

Idle clock: 1506 MHz, stable at 0.7 V

Max boost clock: 1898 MHz, achievable with about +160 MHz at 0.9 V

Max memory clock boost: about +600 MHz

Software used: MSI Afterburner (OC/UV), HWInfo64 (Monitoring)

 

 

The background to this question:

Spoiler

 

I am currently undervolting/overclocking my GTX 1070. This is my first time really min maxing everything, so there is still a lot for me to learn, please be gentle. I also have another thread regarding this "Journey", but considering other people might find this thread during their research in the future and the other is already pretty long, I want to keep it short and easily accessible, hence a new thread.

 

The issue I am running into is that I am unsure what the ideal voltage/frequency combination for my purposes would be, so I am doing a lot of testing. This includes identifying the max stable frequencies for the voltages between 0.9 and 1.05 V. There's 16 steps between these voltages and since I usually have to run through multiple stability tests per step, I want to keep it efficient.

 

 

[Levent]

Because not every load is the same. CPU overclock testing is somewhat straight forward. GPU overclock testing is different and difficult. "Stability testing" and "quickly" wont happen together.

[DobertRownySr]

9 minutes ago, kriegisyourbae said:

Hi all,

 

Which tools/settings are best to identify unstable settings in the shortest amount of time possible?

 

Now before you go "google is your friend", please check the following so I can explain myself.

I have so far been using the following tools for stress testing as well as benchmarking:

-Unigine Heaven

-Unigine Valley

-Unigine Superposition

-OCCT

-Furmark

-3D Mark Fire Strike

-3D Mark Time Spy

-Horizon Zero Dawn (Custom settings)

-Metro Exodus (Ultra Preset)

-Rise of the Tomb Raider (Maxed out settings)

 

With none of the tools/game benchmarks used I was able to find reliable results. While all of them were stable at +180 MHz core/+620 MHz memory clock @ 0.912 V, actually playing RotTR resulted in the game crashing. So I lowered the OC slightly and had to repeat this until I was at +160 MHz core boost. The game had crashed up to 3 hours into playing it. Now playing a game for 2-3 hours for each individual value would be insane, that's easily 100 hours with the main purpose of stress testing. So I want to find a method that's more efficient.

 

That's where my question comes in.

 

 Some general information regarding my scenario:

  Reveal hidden contents

 

Idle clock: 1506 MHz, stable at 0.7 V

Max boost clock: 1898 MHz, achievable with about +160 MHz at 0.9 V

Max memory clock boost: about +600 MHz

Software used: MSI Afterburner (OC/UV), HWInfo64 (Monitoring)

 

 

The background to this question:

  Reveal hidden contents

 

I am currently undervolting/overclocking my GTX 1070. This is my first time really min maxing everything, so there is still a lot for me to learn, please be gentle. I also have another thread regarding this "Journey", but considering other people might find this thread during their research in the future and the other is already pretty long, I want to keep it short and easily accessible, hence a new thread.

 

The issue I am running into is that I am unsure what the ideal voltage/frequency combination for my purposes would be, so I am doing a lot of testing. This includes identifying the max stable frequencies for the voltages between 0.9 and 1.05 V. There's 16 steps between these voltages and since I usually have to run through multiple stability tests per step, I want to keep it efficient.

 

 

I use Unigine Heaven and 3D mark.

Games I test are Cyberpunk 2077, Deep Rock Galactic, and Apex Legends.
Mainly because it seems those 3 games seem more sensitive to OC related crashing.

[kriegisyourbae]

Just now, Levent said:

Because not every load is the same. CPU overclock testing is somewhat straight forward. GPU overclock testing is different and difficult. "Stability testing" and "quickly" wont happen together.

Well but what in your opinion would be the most efficient way. Maybe I shouldn't say quickly as it implies "stability check in 5 minutes". I am just looking for quicker ways than those I described. They're just painful. I don't want to play RotTR for 100+ hours lol

[Levent]

Just now, kriegisyourbae said:

Well but what in your opinion would be the most efficient way. Maybe I shouldn't say quickly as it implies "stability check in 5 minutes". I am just looking for quicker ways than those I described. They're just painful. I don't want to play RotTR for 100+ hours lol

For CPU overclock, Prime95 (Blend test, atleast 6 hours). For GPU overclock, 3Dmark stability test for initial testing, then furmark to check if worst case scenario temperatures are under control then just playing different games for a week.

[DobertRownySr]

1 minute ago, kriegisyourbae said:

Well but what in your opinion would be the most efficient way. Maybe I shouldn't say quickly as it implies "stability check in 5 minutes". I am just looking for quicker ways than those I described. They're just painful. I don't want to play RotTR for 100+ hours lol

I usually run Heaven and gradually change my voltage curve and I push it as far as I can until it crashes.

Then After it's stable in heaven, I test it in 3DMark. Once it's stable in those, then I move onto games. Most games run fine, but some games (like the ones I mentioned above) might crash. If they do, I lower my max clock speed by 15mhz and try again.

[Quackers101]

"quickly" but then its only stable for a couple of days.

[kriegisyourbae]

@DobertRownySr @Levent When you guys talk about 3DMark - are you talking about a premium feature or just the general features available in the free demo?

[DobertRownySr]

1 minute ago, kriegisyourbae said:

@DobertRownySr @Levent When you guys talk about 3DMark - are you talking about a premium feature or just the general features available in the free demo?

I just like 3DMark as a software. It has many different benchmarks, and it seems to push a card harder than heaven (so it helps to weed out stability issues)

[Levent]

4 minutes ago, kriegisyourbae said:

@DobertRownySr @Levent When you guys talk about 3DMark - are you talking about a premium feature or just the general features available in the free demo?

I was talking about 3dmark Stress Test feature. I dont know if its paid or not, I bought 3dmark ages ago.

[kriegisyourbae]

32 minutes ago, Levent said:

I was talking about 3dmark Stress Test feature. I dont know if its paid or not, I bought 3dmark ages ago.

Thanks for the advice, I'll get the full version and report back if there's any noticeable difference compared to other benchmarks/stress tests I used.

[kriegisyourbae]

20 hours ago, Levent said:

I was talking about 3dmark Stress Test feature. I dont know if its paid or not, I bought 3dmark ages ago.

 

20 hours ago, DobertRownySr said:

I just like 3DMark as a software. It has many different benchmarks, and it seems to push a card harder than heaven (so it helps to weed out stability issues)

 

I've bought 3D Mark Advanced yesterday and tested the Fire Strike Extreme Stress Test. It has so far been the only tool that recognized +165 MHz at 0.912 V to be unstable. All other tools had no issues with these values, but RotTR crashed 2-3 hours into a gaming session. This was great advice, thank you guys!

 

I'll do more testing and hopefully be done with all this a lot sooner than I was afraid of.

[kriegisyourbae]

3D Mark Advanced has proven to be very reliable. With the Fire Strike Extreme Stress Test any unstable overclocks have been identified, even though the tools from my initial post didn't detect them. My undervolted GPU is running on a very stable overclock now. If you're interested in the results, here's a screenshot. It compares Stock to both my old, non-undervolted overclock and the new and undervolted overclock. Although I am getting a slightly lesser boost in performance, the temperature and power draw differences are totally worth it.

 

Quick info for context: I am running a GTX 1070 and a Ryzen 5 1600x and used HWInfo64 for monitoring as well as MSI Afterburner for overclocking/undervolting.

 

 

Thanks again for hinting me towards the full version of 3D Mark, that purchase has been absolutely worth it.

 

 

I'll mark this message as the solution for visibility should others be looking for the same thing. Credits go to Levent and DobertRownySr.