Prime95 has stopped working

[PowerK]

I left my PC for Prime95 (v27.7) In-place Large FFT (Maximum heat, power consumption etc..)(the middle option) last night.

I was greeted with "Prime95 has stopped working" message when I returned to PC this morning. Sad thing is, previously, this PC passed 20 passes of LinX and 10 hours of Prime95 small FFT test.

Personally, I've only seen BSOD, freeze and/or rounding errors when Prime95 when it comes to overclock instability.
Anyway, I'm sure "Prime95 has stopped working" message is a sign of instability.

Ok.

I'm using AsRock Fatality Z77 Professional.
3770K @ 4.5 with 4 x G.SKILL ARES 2133MHz (16GB) (1.65V)

Offset mode = +0.050V
Load-line calibration = Auto (Level 5 which is the lowest/mildest)
All power management options are at default (enabled) (C1E, C3/C6 etc)
I have not touched PLL nor VCCSA etc. (Pretty much all these are at default).
Under Prime95 small FFT load, CPU-Z reports 1.200V. And under Prime95 large FFT load CPU-Z reports 1.208V

What should I touch to get prime95 stability ?

 

[Ken g6]

4 x G.SKILL ARES 2133MHz (16GB) (1.65V)

That could be the problem. Large FFT tests memory as much as it tests the CPU. You shouldn't be running 1.65V RAM with Sandy/Ivy Bridge processors. I'd take it down to 1600MHz/1.5V and see if that helps.

 

[tigersty1e]

i would run it again, but do a custom blend with 90% of ram.

it takes 20.5 hours for a full loop with 15 minute tests.

 

[PowerK]

That could be the problem. Large FFT tests memory as much as it tests the CPU. You shouldn't be running 1.65V RAM with Sandy/Ivy Bridge processors. I'd take it down to 1600MHz/1.5V and see if that helps.

The G.SKILL memory is rated to run at 2133MHz @ 1.65V with 9-11-10-28 2T timings
http://www.gskill.com/products.php?index=489

I've used this memory without any problems on Sandy Bridge system.

 

[borisvodofsky]

The G.SKILL memory is rated to run at 2133MHz @ 1.65V with 9-11-10-28 2T timings
http://www.gskill.com/products.php?index=489

I've used this memory without any problems on Sandy Bridge system.

yea dude, it's fine, many people here are just the cautious type th's all. :biggrin:

basically m0ar voltage is the solution to Every overclocking problem once you rule out "heat" :awe:

 

[BD231]

Pretty low volts for that oc, and typically lower numbers suggest a lower LLC, 1 being the lowest. Are you sure level five is the lowest setting? LLC can cause instability in itself if its to high but it can also cause instability overclocking if its to low. Aim for 75% LLC, from there finding your required volts is easy. My 2133 is stable at cas 11, never tried for cas 9 though.

 

[PowerK]

Pretty low volts for that oc, and typically lower numbers suggest a lower LLC, 1 being the lowest. Are you sure level five is the lowest setting? LLC can cause instability in itself if its to high but it can also cause instability overclocking if its to low. Aim for 75% LLC, from there finding your required volts is easy. My 2133 is stable at cas 11, never tried for cas 9 though.

Yup. On AsRock board, Level 5 is the lowest/mildest LLC (Level 1 being the highest/strongest value for LLC).

I let my main PC run the Prime95 v27.7 again (without changing anything for the time being). I selected the In-place large FFTs (maximum heat, power consumption) again just for the heck of it.

So far, the PC is running the Prime95 without any errors for 6 hours and counting.

This is what drives me nuts. If there's an error or app crash due to overclock instability, I want them to be constant under certain scenario. Now, it's running fine even though I did not change any settings.

I'll try custom blend with 90% of my system RAM with FFT 1344 min/max (as I heard this really stresses the CPU and memory sub-system) later tonight.

 

[dma0991]

How stable do you need it to be in the first place? I haven't tested mine more than 2 hours of Prime95, not because it couldn't, I didn't want to or it has met my expectations of what I consider stable enough. Oddly enough most BSODs only happen to me during idle loads or minimal load. Perfectly fine at max load for F@H ~ 7-8 hours.

 

[BD231]

Well you're darn close if it's running over 6 hours in prime, especially if it didn't BSOD. I'd stop putting an overclocked chip through so much stress testing though. No sense in speeding up degradation as there are better ways to feel a chip out.

If its a memory/memory controller error you'll likely see it somewhere else, like gaming or video editing so that's a decent enough test for memory. You can also test to see if your chip is voltage starved by running cinebench and comparing your scores at different voltages. Once your scores stop going up by adding voltage you know you're right within your target area.

 

[PowerK]

How stable do you need it to be in the first place? I haven't tested mine more than 2 hours of Prime95, not because it couldn't, I didn't want to or it has met my expectations of what I consider stable enough. Oddly enough most BSODs only happen to me during idle loads or minimal load. Perfectly fine at max load for F@H ~ 7-8 hours.

That is indeed a good question.

I think if we encounter BSOD during daily usage (light load such as internet, video.. and heavy load such as playing games like Crysis, BF3, MP3 etc), it's not that hard to troubleshoot.

However, app crash is more difficult to troubleshoot (at least, I think so). I'm sure Prime95 app crash is 99% due to overclocking instability. What about.. app crash during playing games and/or during using daily applications? Aren't you always going to wonder whether it's due to overclocking? I don't know.. it may just be me. Maybe I should not overclock at all.

Well you're darn close if it's running over 6 hours in prime, especially if it didn't BSOD. I'd stop putting an overclocked chip through so much stress testing though. No sense in speeding up degradation as there are better ways to feel a chip out.

If its a memory/memory controller error you'll likely see it somewhere else, like gaming or video editing so that's a decent enough test for memory. You can also test to see if your chip is voltage starved by running cinebench and comparing your scores at different voltages. Once your scores stop going up by adding voltage you know you're right within your target area.

Thank you for your insight! Appreciated.

 

[Brunnis]

How stable do you need it to be in the first place? I haven't tested mine more than 2 hours of Prime95, not because it couldn't, I didn't want to or it has met my expectations of what I consider stable enough. Oddly enough most BSODs only happen to me during idle loads or minimal load. Perfectly fine at max load for F@H ~ 7-8 hours.

There's no such thing as "stable enough". It's not the occasional crash that's the real issue, it's the silent data corruption that can occur. I personally run several different stability tests for ~24h each and I usually do other things as well while they're running (Prime 95 + Firefox internet browsing has uncovered errors quite fast for me in the past).

 

[dma0991]

Aren't you always going to wonder whether it's due to overclocking? I don't know.. it may just be me. Maybe I should not overclock at all.

We can't assume that you've passed x hours of Prime95 would mean that the system is rock solid, even capable of handling mission critical applications. There are definitely scenarios that Prime95 may not test, and when that happens, instability. I run a mix of Prime95, IBT, LinX and OCCT to find the nearest possible stable overclock.

You're running the CPU out of its factory specifications, crashes are bound to happen, unless you happen to fall in the category of overclocked and stable for years.

For my daily use, I find there is sufficient processing power and 4.5GHz didn't make any difference, so I decided to use stock settings most of the time. Whenever I run F@H, I'll overclock it as it shaves off at least 1-2 hours for each work unit. It is only for that particular usage that I find overclocking gave me some benefit over stock.

There's no such thing as "stable enough". It's not the occasional crash that's the real issue, it's the silent data corruption that can occur. I personally run several different stability tests for ~24h each and I usually do other things as well while they're running (Prime 95 + Firefox internet browsing has uncovered errors quite fast for me in the past).

Of course there is, it depends on what you need it to do. Not quite sure why are you running Prime95 and browsing at the same time, I would usually leave it untouched for the duration it is running Prime95.

 

[Brunnis]

Of course there is, it depends on what you need it to do.

Nope, there's no middle ground. Just because it works with one game or applicaton, doesn't mean it will work with another.

Not quite sure why are you running Prime95 and browsing at the same time, I would usually leave it untouched for the duration it is running Prime95.

Because the additional stress has proven effective in uncovering some stability issues for me in the past. Do you really think that if Prime95 + Firefox crashes, that's the ONLY case where the CPU would fail? One failing case should be enough for anyone that values their data to reevaluate their overclock.

 

[dma0991]

Nope, there's no middle ground. Just because it works with one game or applicaton, doesn't mean it will work with another.

Because the additional stress has proven effective in uncovering some stability issues for me in the past. Do you really think that if Prime95 + Firefox crashes, that's the ONLY case where the CPU would fail? One failing case should be enough for anyone that values their data to reevaluate their overclock.

Why wouldn't there be a middle ground. It is a fair tradeoff for me to find the lowest possible voltage to sustain my overclock. Surely it wouldn't be rock stable in every situation, but what I lose in stability I gained in longevity. Surely I could push a few more volts to make it stable but for what I do with my PC, I find it unnecessary for that small difference.

It makes no sense that I should run Prime95, IBT and LinX at the same time just so that it creates more stress than it would for me to run them one after the other. Each one of them takes up all of your resource, adding an extra program that would also need some memory space and CPU time is just unnecessary.

Maybe I should play Skyrim as I run Prime95? Certainly it wouldn't be considered stable unless I managed to do both successfully? :hmm:

 

[PowerK]

Ok.

Passed another 12 hours of Prime95 In-Large FFT (maximum heat, power consumption option).

Played a few hours of Max Payne 3. Looked good. No BSOD, no app crash. So I thought everything was fine.

Until.... I checked Windows system logs..... There were several WHEA errors listed/logged while I was playing the game. I'm back to stock (3.5GHz)

 

[2is]

Could be a problem with Prime. I had the same issue with prime where it would stop working at the exact same spot every time regardless of settings. I deleted and downloaded another version, all was well after that.

 

[BD231]

Could be a problem with Prime. I had the same issue with prime where it would stop working at the exact same spot every time regardless of settings. I deleted and downloaded another version, all was well after that.

His LLC is to low and nobody has said much about stability at cas 9 running 2133 memory, of course AS rock boards have auto tuning so he could just set it to 4.4ghz via the Auto Tune feature if he can't figure it out himself.

 

[Puppies04]

Why wouldn't there be a middle ground. It is a fair tradeoff for me to find the lowest possible voltage to sustain my overclock. Surely it wouldn't be rock stable in every situation, but what I lose in stability I gained in longevity. Surely I could push a few more volts to make it stable but for what I do with my PC, I find it unnecessary for that small difference.

Read his previous post and go look up "what silent data corruption" is, that is why there is no middle ground. This isn't the best link ever but i'm sure it will highlight the point he is trying to make.
http://en.wikipedia.org/wiki/User:CTho/Overclocking

 

[PowerK]

An interesting thing is that I get the WHEA error only when I play Max Payne 3.
Max Payne 3 must be very CPU intensive game.

3770K @ 4.5
Offset +0.050V
LLC = Level 5 (lowest/mildest)
16GB 2133MHz RAM @ 1.65V with 9-11-10-28 2T
Everything else at Auto.

- Passed 10 hours of Prime95 v27.7 (In-large FFTs and custom blend with 90% of RAM)
- Passed 20 passes of LinX
- No BSOD, no app crash.
- Played Crysis and Serious Sam 3 BFE (and Diablo III as well. But Diablo III is such an easy game for CPU) fine without any errors.

However...
- While playing Max Payne 3, I can see WHEA error in Windows system log. (event viewer)
Hence, I tried to raise offset to +0.060V. The same result. WHEA errors during Max Payne 3 play. And with +0.060V temps were too high. (almost 95C).

Unless I manually checked Windows System Log in Event Viewer, I'd have never known if my overclock was really stable or not... because CPU stress tests are all passed, plays games and others just fine without any noticeable stability issues such as BSOD, app crash etc.

Grrr.... what to do..

 

[2is]

Solution Sounds pretty simple. Lower your OC to 4.4 you aren't going to miss that 100 MHz. ePeen pride might take a hit but that's all.

 

[PowerK]

Solution Sounds pretty simple. Lower your OC to 4.4 you aren't going to miss that 100 MHz. ePeen pride might take a hit but that's all.

All right. I'll try 4.4 and report back.

 

[PowerK]

Ok.

At 4.4GHz,
Offset mode = +0.010V
LLC = Level 5 (lowest/mildest)
16GB 2133MHz RAM @ 1.65V with 9-11-10-28 2T
Everything else at Auto.

CPU-Z & AIDA64 CPUID report 1.160V under LinX load. (quite a bit lower than what it required at 4.5 which was about 1.200V ~ 1.208V)

- I played Max Payne 3 for a few hours without WHEA errors.
- 20 passes of LinX without BSOD, app crash nor WHEA errors.

So far so good.

This is my first time trying 4.4 with this 3770K. But wow.. it seems there's a stiff wall between 4.4 and 4.5 Vcore and temp difference is surprising.

Started Prime95 custom blend test with 90% of my RAM (13,500 MB) about an hour ago.

Will post again after about 11~12 hours.

 

[JAG87]

Pretty normal results...

I'm at 4.3 wtih 1.080V under load (+0.005 offset, I have lower stock voltage than most). You're at 4.4 with 1.160V, which is more or less what I need for 4.4 (+0.060V offset if I recall correctly), and 4.5 needed +0.110 (which was 1.190V). But these weren't fully tested, just basic sanity checks, so I wouldn't be surprised if stability required voltages very similar to yours.

I didn't find the extra voltage and heat worthwhile for only an extra 100-200 Mhz.

 

[PowerK]

Bah... WHEA errors @ 4.4 with +0.010V offset. (during Prime95 custom blend).
I don't want to raise offset. I'll try 4.3 with +0.005V offset.

JAG87,

No WHEA errors on your end 4.3 with +0.005V offset ?

 

[2is]

I needed +.015 at 4.2 to be fully stable. LLC off though.