PC is still unstable after testing with stress test

[The_Misanthrope]

I've been trying to overclock my GPU. I've been testing the settings by using Kombuster to check the stability. I run the program with these settings and it has never crashed, I have even tested it for half an hour without any crashes and the temperature doesn't go over 82C. But when I do normal computer stuff with these settings or play games, the computer crashes and I have to restart it. If my computer is unstable with these settings, why doesn't it crash under 100% load when testing?

 

My power supply is 500W

[Mark Kaine]

Because the test is bad.

 

 

Try heaven. 

[Vishera]

6 minutes ago, The_Misanthrope said:

I've been trying to overclock my GPU. I've been testing the settings by using Kombuster to check the stability. I run the program with these settings and it has never crashed, I have even tested it for half an hour without any crashes and the temperature doesn't go over 82C. But when I do normal computer stuff with these settings or play games, the computer crashes and I have to restart it. If my computer is unstable with these settings, why doesn't it crash under 100% load when testing?

 

My power supply is 500W

Stress tests only test power consumption and heat,they don't test stability.

 

But you can try testing the stability of the GPU with GPUPI 3.3.3,Press Calculate then use the following settings:

32B HWbot GPU

choose your GPU in OpenCL GPU devices

Batch size: 20M

Reduction size: 512

Then press OK to start.

(Press cancel if a message about HPET appears)

 

Tell us if you get any errors while testing.

 

GPUPI 3.3.3 download page:

https://www.overclockers.at/news/gpupi-international-support-thread

[Vishera]

1 minute ago, Mark Kaine said:

Because the test is bad.

 

 

Try heaven. 

GPUPI is basically the Prime95 of GPUs,just with relatively less heat and power consumption 

[--SID--]

What brand and model GPU and PSU?

[jones177]

My FTW3 Ultra 2080 ti could do +175 on the cores with Fire Strike, +150 on the cores in Time Spy but no more than +117 in non RT games and only +110 in RT games.

Heaven was good for +120 so it is the closest bench to games. 

 

With my EVGA GTX 1080 ti SCs any overclock on the cores crashed most games so it was memory only.

 

With the 30 series cards I don't know because non have crashed yet and all have been tested with a +100 on the cores and +800 on the memory.

 

No card will perform well at 82c so you may want to get the heat under control first.

 

 

[Dr_badwolf]

42 minutes ago, The_Misanthrope said:

I've been trying to overclock my GPU. I've been testing the settings by using Kombuster to check the stability. I run the program with these settings and it has never crashed, I have even tested it for half an hour without any crashes and the temperature doesn't go over 82C. But when I do normal computer stuff with these settings or play games, the computer crashes and I have to restart it. If my computer is unstable with these settings, why doesn't it crash under 100% load when testing?

 

My power supply is 500W

This is a memory callback error. It will work most of the time, but now and then it messes something up too much and turns into Dory (the fish, ya)

[The_Misanthrope]

52 minutes ago, Vishera said:

Press cancel if a message about HPET appears

When I started the program, it opened a message that said something about an "unsafe timer". Is that what you meant? I saw the message and just closed the program and my system just restarted by itself.

[Dr_badwolf]

4 minutes ago, The_Misanthrope said:

When I started the program, it opened a message that said something about an "unsafe timer". Is that what you meant? I saw the message and just closed the program and my system just restarted by itself.

HPET is a... thing that measures time "most exactly" when doing something that requires measuing time.
A Simulation, for example.
A simulation in physics requires [time] in its calculation. which always needs HPET.
Game physics dont care "as hard" because youre not about to launch a rocket with Doom.
CUDA Physics engines will test your GPU in 'another way' So go for it.

You can turn off HPET after, it will tax games @ like... 2-5% last i looked it up... 7 years ago. 
 

[Dr_badwolf]

However... Because you say it doesnt go over reasonable temps and still crashes, this makes me think the voltage is "tu lo"
If your crashing below temp specs, inch up your Volts.
So long as you do it...
- an "itty bit" at a time.
- Crashes
- Reset
- more V
- etc... 
it will be relatively safe to bump it up.
So long as you keep it under the Max Watts of the cards power.
(98% safe. Its what/how i do.)

[The_Misanthrope]

I'm trying to overclock without increasing the voltage. I'm trying to get the most performance out of stock voltage.

[Mark Kaine]

1 hour ago, Dr_badwolf said:

98% safe. Its what/how i do.)

My card is rated 270w,  draws easily 275 or so, with undervolt... pretty sure Nvidia cards automatically know the limits, regardless of what user does,  unless they mod bios/card.

 

52 minutes ago, The_Misanthrope said:

I'm trying to overclock without increasing the voltage. I'm trying to get the most performance out of stock voltage.

And that's fine, depending on "silicone lottery" more voltage might wouldn't even work. 

Idk what's up with this hpet error tbh, but just use heaven benchmark... if oc is unstable it will crash very quickly...

 

You can set it windowed 720p, that way you can do adjustments in real time, or also just check voltages,  clocks, temps, etc in afterburner...

 

Ex:

Spoiler

 

In my experience like if its stable in heaven it'll be stable in almost every game.  

Also its a good idea to set it 10mhz lower so you aren't at the absolute limit of crashing all the time, after you found a "stable" oc in heaven.

 

[Dr_badwolf]

4 minutes ago, Mark Kaine said:

My card is rated 270w,  draws easily 275 or so, with undervolt... pretty sure Nvidia cards automatically know the limits, regardless of what user does,  unless they mod bios/card.

 

And that's fine, depending on "silicone lottery" more voltage might wouldn't even work. 

Idk what's up with this hpet error tbh, but just use heaven benchmark... if oc is unstable it will crash very quickly...

 

You can set it windowed 720p, that way you can do adjustments in real time, or also just check voltages,  clocks, temps, etc in afterburner...

 

In my experience like if its stable in heaven it'll be stable in almost every game.  

Also its a good idea to set it 10mhz lower so you aren't at the absolute limit of crashing all the time, after you found a "stable" oc in heaven.

 

Im going to try to explain something very specific in my next post (cause more to this would be messy)...

[Vishera]

1 hour ago, The_Misanthrope said:

When I started the program, it opened a message that said something about an "unsafe timer". Is that what you meant? I saw the message and just closed the program and my system just restarted by itself.

Yes,just press no when it asks that.

[Dr_badwolf]

Ill try to teach this once.
BECAUSE... perhaps im not good at explaining it.

[Dr_badwolf]

feel free to ask for clarification where i was confusing.
Happy to help - but its more your patience than mine. 

 

[Mark Kaine]

23 minutes ago, Dr_badwolf said:

feel free to ask for clarification where i was confusing.
Happy to help - but its more your patience than mine. 

 

While what you posted isn't wrong,  its more for cpus, because of the whole voltage stuff, with gpus you can basically just ignore that and simply inch your way forward...

 

Core clock 25mhz steps.

 

Then memory,  50mhz steps.

 

 

Always test with heaven after each step... if you have a bit of experience you don't even need a full run... let it run 1 minute,  if no crash, next step...

 

You can do full runs once a supposed stable oc was found. 

 

 

That said,  the way most people overclock is weird to me, because they always run against a voltage/ temp limit... it's far easier to undervolt,  usually you get lower temps and thus higher, sustained clocks... thats at least true for many 2000 and 3000 cards .

 

 

[Vishera]

50 minutes ago, Mark Kaine said:

While what you posted isn't wrong,  its more for cpus, because of the whole voltage stuff, with gpus you can basically just ignore that and simply inch your way forward...

 

Core clock 25mhz steps.

 

Then memory,  50mhz steps.

 

 

Always test with heaven after each step... if you have a bit of experience you don't even need a full run... let it run 1 minute,  if no crash, next step...

 

You can do full runs once a supposed stable oc was found. 

 

 

That said,  the way most people overclock is weird to me, because they always run against a voltage/ temp limit... it's far easier to undervolt,  usually you get lower temps and thus higher, sustained clocks... thats at least true for many 2000 and 3000 cards .

 

 

Clock drifting is an issue even with GPUs,

That's why HWBot requires you to bench GPUPI with HPET enabled.

[Mark Kaine]

31 minutes ago, Vishera said:

That's why HWBot requires you to bench GPUPI with HPET enabled.

T b h, i don't know what HPET does... its a windows clock thing, but what does it do?

 

Edit: ok I just checked,  i have it on, I do not cheat!!! (btw did you know you can cheat gpu benchmarks by setting texture filtering to performance and other shenanigans ... )

 

Firestrike even "invalidates" the results if you do it too much. : D

 

Ps: i always have texture filtering on quality... except on my laptop...

 

 

[Vishera]

33 minutes ago, Mark Kaine said:

T b h, i don't know what HPET does... its a windows clock thing, but what does it do?

High Precision Event Timer - As far as i know it corrects clock drifting,

but in many situations it has slower performance than RTC - Real Time Clock.

[The_Misanthrope]

I just tried the GPUI program and it aborted and said that the result was invalid. The page says that happens if the drivers are old, but I just checked for new versions and my driver is the latest.

 

Nevermind, I reduced the reduction size. What do the results tell me?

[The_Misanthrope]

On 11/1/2021 at 4:43 PM, Dr_badwolf said:

This is a memory callback error. It will work most of the time, but now and then it messes something up too much and turns into Dory (the fish, ya)

Yeah, I think that's it because whenever I increase the memory clock even slightly from stock settings, the Heaven benchmark crashes.

[Dr_badwolf]

On 11/4/2021 at 12:01 PM, The_Misanthrope said:

Yeah, I think that's it because whenever I increase the memory clock even slightly from stock settings, the Heaven benchmark crashes.

Look into GPU Bios editors.
GPU Ram is the same as DIMMS.
Just the settings are harder to get to.