Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Sandy B Ridge]

After the recent discussions about broken PCs and broken PC bits, I thought I'd ask you clever guys what methodology do you use to stress test your setup. Memory, CPU, Graphics card, Cooling, temperatures etc. etc.

Which programs do you use? Do you have a set way of doing it?

Cheers,

SBR

[Hendrik]

There isn't really "one" tool to do it all properly, that i know of anyway.

I use prime95 to do the usual CPU torture testing, which then also includes thermal testing. Can use any thermal monitor of your choice to monitor the temps at the same time.
I usually run that on a newly setup PC for a short while to see its thermal properties.

If i suspect memory problems, memtest86 is still one of the best tools, imho.
I don't usually test a working PC though, only when i suspect something may be off.

[InflatableMouse]

Be prepared for long hours possibly days waiting and unable to use your pc. Also, be prepared to deal with issues as they might arise. Memory issues, bluescreens etc. Make backups before you begin as a bluescreen may corrupt your startup disk (its unlikely but can happen).

Read up! Go to overclock.net and read guides on stress testing. Whatever tests people run to get an overclock stable is applicable to a stock machine too. Just don't apply their fixes as it may not apply to a stock machine. Testing is the same. If you're not comfortable about it, if its too technical don't do it!

Very important note: stress testing may permanently damage parts as the heat generated can be extreme. With stock coolers and/or bad case ventilation some parts may get too hot and cause permanent damage. Imagine heated air is not ventilated out of your case properly. Each degree the air rises in your case, is fed back to the CPU and causes your CPU to heat up equally. This effect keeps itself going! Stress testing also stresses your PSU, it may fail or even burn out. Its all not very likely to happen but consider yourself warned.

I go through the below on every new pc or when I change parts like CPU, motherboard or memory. Also, I run this every once in a while (6 months to a year) as variables can change.

Document all your current BIOS settings and then reset the BIOS to optimized defaults and make necessary changes like AHCI and boot order. Make sure memory timings are correct (preferably use SPD as it reads the manufacturer's timing tables from the memory itself). At this point, do not overclock or change anything else, you want the most basic, default settings to make sure everything is oke. DO NOT OVERCLOCK! Even if you want to test overclock stability, its meaningless if you don't know if you're machine is stable stock, you need a stable basis for overclocking.

Start with memtest86+, not to be mistaken with memtest86, which isn't free and (although the original program) not nearly as good as the + version. Memtest is harmless as far as I'm concerned and can be left over night.

Memtest can be downloaded as a CD image or isntalled to a USB memory stick. Whatever suits you best. Boot from it and memtest+ starts automatically.

Let it finish, a full cycle takes several hours. If you can, let it run for 2 or 3 cycles (I leave it over night). You need to run memtest before anything else because if issues arise during CPU stress testing, resulting in a bluescreen or something, you won't know 100% certain if it could be memory related. Errors will be shown with addresses.

If it finishes fine you're ready for CPU stress testing.

Download the following:
CPU-Z
Real Temp
Prime95 - pick the version for your Windows, v2.66 is the latest. Doesn't require install, just unpack.

Open CPU-Z and RealTemp and let your PC idle for a few minutes and take note the core voltage and frequency in CPU-Z and of the temperatures in Real Temp. If that looks fine, nothing weird continue and open settings in Real Temp and enable log file 5 secs. The log file will be in the program folder where its installed. If that disk doesn't have a lot of free space, you may want to install the program on another drive as 24h of logging can get large . You can use the log to possibly see what happened in case of a crash/reboot if you weren't there. Also under settings, enable the alarm and use a max temperature you're comfortable with. For me thats 80C but different CPU's have different max temperatures.

If you're on Win7, enable all icons in the notification area. Prime has the nasty habit of minimizing to the notification area when you close it. If your cpu is overheating and you want to close prime, it will still be running! You need to choose Test/Exit or Test/Stop and in the popup, stop all workers.

Also, if you can, make sure all fans are running at maximum for the first time as a safety measure.

Start Prime95 and run the Blend test (leave all default). Watch Real Temp like a hawk! Distance to TJ Max is supposed to be the theoretical maximum your CPU can rise before reaching its critical temperature. I would advise to try and stay at least 10 degrees below that. To give you an idea, my TJ Max is 98C. I stop Prime if it gets over 80C. For day to day use, I want CPU temps to stay below 70C.

Next to watching Real Temp, keep an eye on the Core Voltage in CPU-Z. It should not drop or fluctuate. Even as much as a 100th of a volt can cause instability. <-- read the replies below, on new systems with default settings this should no longer be an issue.

Several things can happen when you run Prime:
It runs into infinity without an issue, all workers have a green icon in Prime and your temperatures are fine. To be quite sure your hardware is stable, run it for 12h minimum. To be absolutely sure, run it for at least 24h. There have been several issues with Prime failing at 16-18 hours.
It bluescreens in which case you need to make note of the code, 0x00000101 -or something.
A worker fails, it will get a red icon in Prime. This means your CPU made a calculation error. Not good.
PC hangs, no mouse/keyboard and requires a hard reset.
Runs fine, no faults or crashes but temperatures are out of control. Stop Prime immediately and improve your cooling .

I can't tell you how to fix crashes/bluescreens/hangs. I'd need to know your settings, hardware and what happened and even then its hard without actually seeing/doing it myself. If you decide to go ahead, make sure you've done your homework, read up!

One more note on Prime: after a few minutes the temperatures will LOOK stable! Don't be fooled! Often it starts to rise again after 30 minutes or an hour! If temps are stable after a few hours, you can leave it alone. For the first few, personally I am always staying around ready to intervene!

And whatever you do, don't panic and pull the plug! If you must panic, do a reset . If you pull the plug, all fans stop and your CPU will be left shimmering in its own heat. Reset it and all fans will keep running but CPU load is removed. Of course try to stop Prime manually is preferred though .

Last but not least, IMHO, if you're not prepared to do the full runs of each test, don't bother doing them at all. There is no point in running memtest for 1 hour and prime for 2 if your intention is to stress test and test stability.

Oh and uh ... don't look at me if things fail/break/burn .

Good Luck!

[JimH]

Wow!  Thanks for taking the time to write this up, InflatableMouse.  Great job.  I'll add it to the wiki.

[InflatableMouse]

Cool  .

Didn't realize that it would be wiki material as its rather offtopic as far as MC is concerned. I might rewrite/reformat it to a more readable, structured article.

[glynor]

That's essentially my process too, InflatableMouse.  Though, I don't typically run Memtest86+ for more than 1-2 passes anymore.  I find it isn't a very good test of general stability, and failures are very rarely found after the fist run which wouldn't be more quickly found by running Prime once you're booted.  I do have one question and one or two additional tips though...

Why do you prefer RealTemp over CoreTemp?  Just curious if there is a reason I should switch, since I'll be dancing this dance soon here.

As far as tips, I have three additional things:

1. It is a VERY GOOD IDEA to do this whole process using a fresh Windows installation that you don't care about, and without additional disks installed.  If you do run into trouble and Prime95 causes your PC to lock up or crash, you can corrupt your data on disk.  I've had it hose Windows installations before.  The way I do it is that I have a separate "testing boot drive" with just Windows, the testing tools I need, and the drivers installed for the PC.  I unplug my "real" hard drives, and plug in just the testing disk and boot to that during testing.

2. Inflatable's testing regime doesn't test GPU stability much at all.  I've found machines that can pass the Prime95 stress test with flying colors, but which crash or fail when you try to run a 3D game or things like that (usually because of a failing/inadequate Power Supply or overclocking the PCIe channels inadvertently).  For this, I usually run OCCT and/or a single pass of 3DMark as the last step after I've done everything described above.

3. Running Prime for 2-3 hours (which often gets you "one pass" in and should be enough for the temps to stabilize) can, IMHO, be handy as an initial testing tool, even if you don't have time to run the full pass right then-and-there.  I totally agree that this is not "sufficient" testing to confirm stability, as I have seen machines that would fail in the 15-17 hour range.  However, the vast majority of problems manifest within the first hour or so while the cores are heating up.  So, if you have time for a 2-3 hour run, but not the full 24 for whatever reason, it can still be handy to give yourself a sanity check before you make any additional changes or tweaks.  Another handy tool for "quick tests" while you are tweaking the BIOS settings is OCCT, which pushes it hard and pushes it fast.  It can also be used to stress test the GPU so it is a handy tool to have.

Also, like InflatableMouse suggests, I always stress test any new machine at stock, not just when I have trouble or have overclocked it.  You never know.  I've had brand-new machines at stock freak out and overheat because I messed up installing the CPU goo or something like that (once I forgot to plug in the fan header for the heat sink on the CPU).  For this reason, and the ones mentioned previously, it is very important to keep an eye on the temps early on in the process.  The way I usually test Prime is to run it when I get home from work while making dinner or something.  This lets me check on the process regularly, and then eventually I go to bed and let it run the rest of the night (and maybe the next day while I'm at work if it is still going strong).

Lastly, just to reiterate this:

A worker fails, it will get a red icon in Prime. This means your CPU made a calculation error. Not good.

What this means is that your CPU was asked to do a complex calculation where the "correct answer" was known ahead of time, and your CPU gave the wrong answer.  That is very bad.  If you can't trust your CPU to do math properly, its primary job, then you can't trust it at all.

I only mention it because I occasionally see people say things like:  It is a stable overclock, it fails Prime, but only on one core and only after a few hours of testing.  No!  That's the point.  You're driving it hard to force it to expose errors that only happen in certain special circumstances.  If it is failing, then it is doing math wrong.

That is decidedly Not Good.

[Hendrik]

keep an eye on the Core Voltage in CPU-Z. It should not drop or fluctuate. Even as much as a 100th of a volt can cause instability.

An important thing to know is that voltages do scale depending on the load the CPU is under, at least on modern CPUs like Sandy Bridge or Ivy Bridge.
The voltage on idle and on 100% load will be different, so don't freak out if that happens.

[InflatableMouse]

An important thing to know is that voltages do scale depending on the load the CPU is under, at least on modern CPUs like Sandy Bridge or Ivy Bridge.
The voltage on idle and on 100% load will be different, so don't freak out if that happens.

That is probably true. Mine no longer fluctuates from idle to load but that took some configuring and I forgot about that when I wrote that up.

On a stock machine it might do that a little bit when load changes, it depends on the hardware used; some boards/PSU's really do better than others. However, under full load with Prime, it shouldn't continue to fluctuate and at most just drop a little bit, I'm guessing here but probably no more than a few dozen millivolts.

[glynor]

Pssst...

Why do you prefer RealTemp over CoreTemp?  Just curious if there is a reason I should switch, since I'll be dancing this dance soon here.

[InflatableMouse]

lol ..

I was typing a big reply until my browser accidently went back and I lost the entire post (I was almost done  ).

I'm redoing it, but i need to go, like, right now Baseball training with my son.

To Be Continued ...

[glynor]

I was typing a big reply until my browser accidently went back and I lost the entire post (I was almost done  ).

I hate it when that happens.

[InflatableMouse]

That's essentially my process too, InflatableMouse.  Though, I don't typically run Memtest86+ for more than 1-2 passes anymore.  I find it isn't a very good test of general stability, and failures are very rarely found after the fist run which wouldn't be more quickly found by running Prime once you're booted.

Yeh thats fine and why I only mentioned more runs if possible, but 1 complete run is essential. Just don't stop the first run half way. It's also why I always tell people to prepare for long hours even days. If you're gonna do something, do it right .

On another note, Prime is NOT a memory testing tool. It mentions it on that starting screen but I feel its misleading. What it does is it USES memory, but all programs do that as no program can run without memory. The difference though is that Prime has its answers with checksums. Does that really test memory? To some extend, yes. Does it replace Memtest? Never! In fact Prime does one specific test out of the entire suite of tests that Memtest is running.

It's like saying that Memtest tests CPU too, and it does to a certain extend. It uses the CPU because it can't get around not using it. And Memtest knows the answers of its tests too just like Prime. It makes Memtest just as much a CPU testing tool as it makes Prime a memory testing tool.

Why do you prefer RealTemp over CoreTemp?  Just curious if there is a reason I should switch, since I'll be dancing this dance soon here.

I don't and I use both. There was a version of Core Temp once that didn't work well on certain machines (mine in particular ) and that's when I started using Real Temp too. That bug is no longer present though and both programs work equally well I think. The reason why I chose Real Temp in my quick little guide here is that I think its easier to read due to the bigger font on temperature. In addition, Core Temp shows the frequency staticly, it doesn't update it real time as CPU-Z does when Speedstep is enabled. That could lead to questions/confusion. Then there's also VID on Core Temp which is sometimes believed to be another vcore number - confusing too if you don't know what it is. So, to minimize confusion to the OP I'd figured Real Temp was easier.

I love Core Temp though because it shows Watts! DA POWAH! E-Peen! I know, I'm such a geek!

1. It is a VERY GOOD IDEA to do this whole process using a fresh Windows installation that you don't care about, and without additional disks installed.  If you do run into trouble and Prime95 causes your PC to lock up or crash, you can corrupt your data on disk.  I've had it hose Windows installations before.  The way I do it is that I have a separate "testing boot drive" with just Windows, the testing tools I need, and the drivers installed for the PC.  I unplug my "real" hard drives, and plug in just the testing disk and boot to that during testing.

That's what I do too, but am I really going to advise someone to rebuild his rig? 

I figured if he would really proceed and run into issues, he'd post up and someone would be able to see whether its a software fault or a hardware fault.

2. Inflatable's testing regime doesn't test GPU stability much at all.  I've found machines that can pass the Prime95 stress test with flying colors, but which crash or fail when you try to run a 3D game or things like that (usually because of a failing/inadequate Power Supply or overclocking the PCIe channels inadvertently).  For this, I usually run OCCT and/or a single pass of 3DMark as the last step after I've done everything described above.

  You know, I didn't even read about GPU and PSU, I just read CPU/Memory stress test and clicked reply and started typing vigorously.

Personally I use 3D Mark 11, 3D Mark Vantage, Unigine Heaven and a couple of games (like Crysis 2 with hires texture pack and DX11 Patch).

I always run Furmark too just to see how it holds up but I don't swear by it like some others. I don't see Furmark as the Prime95 for GPU's, it doesn't test a GPU like Prime tests a CPU. What Furmark does do well is generate as much heat as possible and in that respect I use it. 3D Mark and good game is a much better test of a graphics card IMO.

3. Running Prime for 2-3 hours (which often gets you "one pass" in and should be enough for the temps to stabilize) can, IMHO, be handy as an initial testing tool, even if you don't have time to run the full pass right then-and-there.  I totally agree that this is not "sufficient" testing to confirm stability, as I have seen machines that would fail in the 15-17 hour range.  However, the vast majority of problems manifest within the first hour or so while the cores are heating up.  So, if you have time for a 2-3 hour run, but not the full 24 for whatever reason, it can still be handy to give yourself a sanity check before you make any additional changes or tweaks.  Another handy tool for "quick tests" while you are tweaking the BIOS settings is OCCT, which pushes it hard and pushes it fast.  It can also be used to stress test the GPU so it is a handy tool to have.

Yeh .. I guess ... hesitatingly. I'm pulling difficult faces here. The OP asked for stress testing methodologodolyody, and that's what I gave. A quick sanity check, although useful and serving a purpose in its own right IF that's what is wanted, is not a stress test.

But let me re-emphasize, 2-3 hours is NOT a stress test, it doesn't 'prove' anything other than that there isn't a major issue. In my experience, if Prime^[1] fails in 2-3 hours you are already having issues, application crashes, memory reference errors or even STOP errors.

_{1. Prime95 v2.66, there's a BIG difference with older versions!}

Also, like InflatableMouse suggests, I always stress test any new machine at stock, not just when I have trouble or have overclocked it.  You never know.  I've had brand-new machines at stock freak out and overheat because I messed up installing the CPU goo or something like that (once I forgot to plug in the fan header for the heat sink on the CPU).  For this reason, and the ones mentioned previously, it is very important to keep an eye on the temps early on in the process.  The way I usually test Prime is to run it when I get home from work while making dinner or something.  This lets me check on the process regularly, and then eventually I go to bed and let it run the rest of the night (and maybe the next day while I'm at work if it is still going strong).

Yup.

Turn on the alarm in Real Temp and give it a conservative threshold. After so-many hours you can leave it alone but the first few hours are crucial. I tend to just sit there and watch it for an hour or so lol. I do browse the internet or do some other minor stuff when Prime runs though.

Before this i7 I'm running now, I very rarely have had a machine that was plug and play 100% stable. Almost every machine I had needed some tuning, even at stock speeds and BIOS defaults. Most common were memory errors. Many motherboards did not detect timings correctly and even if they did, there were compatibility issues that could be resolved with minor memory setting changes (if you've noticed many Asus BIOS revisions make mostly mention of increased memory compatibility). And that's why I love this board so much, I installed it, turned it on, ran memtest and prime and it was 100% stable. No fuss whatsoever!

Lastly, just to reiterate this:
What this means is that your CPU was asked to do a complex calculation where the "correct answer" was known ahead of time, and your CPU gave the wrong answer.  That is very bad.  If you can't trust your CPU to do math properly, its primary job, then you can't trust it at all.

Yes it is bad, and if a stock machine does that you need to RMA the CPU.

Purely from a hardware/stability perspective though this is a "better" situation than a STOP error. A Prime calculation fault is "closer" to being stable than a blue screen is in terms of overclocking and getting it stable. A bluescreen may need a 1/10th of a volt to fix while a "simple" calculation error only requires 1/100th of vcore increase.

I only mention it because I occasionally see people say things like:  It is a stable overclock, it fails Prime, but only on one core and only after a few hours of testing.  No!  That's the point.  You're driving it hard to force it to expose errors that only happen in certain special circumstances.  If it is failing, then it is doing math wrong.

That is decidedly Not Good.

+1.

A fail is a fail and it doesn't matter if its after a minute or 20 hours. A thread fails its bad. Period.

[Hendrik]

That is probably true. Mine no longer fluctuates from idle to load but that took some configuring and I forgot about that when I wrote that up.

On a stock machine it might do that a little bit when load changes, it depends on the hardware used; some boards/PSU's really do better than others. However, under full load with Prime, it shouldn't continue to fluctuate and at most just drop a little bit, I'm guessing here but probably no more than a few dozen millivolts.

It has nothing really to do with the PSU, it just works that way.
The CPUs power regulators will lower the voltage as the CPU is bored.

Mine right now scales from 1.05V (at idle, clocked at 1.6GHz) to 1.35V (load, clocked 4.4GHz).
Even at load it slightly fluctuates between 1.344 and 1.352, as the power regulators adapt.

The times of one fixed voltage are over. I'm sure i could configure my BIOS to behave that way, but there is no advantage in it.

[glynor]

Right with you on everything there and thanks for the explanation on CoreTemp.  I just wondered.  I see some people use one, and others use the other, and I wondered if I was missing something by sticking to CoreTemp.

I only usually "read up" on the current best practices when I get a new machine, which doesn't happen that frequently.

On another note, Prime is NOT a memory testing tool. It mentions it on that starting screen but I feel its misleading. What it does is it USES memory, but all programs do that as no program can run without memory. The difference though is that Prime has its answers with checksums. Does that really test memory? To some extend, yes. Does it replace Memtest? Never! In fact Prime does one specific test out of the entire suite of tests that Memtest is running.

It's like saying that Memtest tests CPU too, and it does to a certain extend. It uses the CPU because it can't get around not using it. And Memtest knows the answers of its tests too just like Prime. It makes Memtest just as much a CPU testing tool as it makes Prime a memory testing tool.

+1

I agree, and testing 1-2 passes in Memtest is certainly part of my routine for this reason.  I just find that the "old way" of running multiple Memtest passes isn't very helpful anymore with modern multi-core CPUs and integrated memory controllers.  I think that is mostly just leftover voodoo from the bad-old-days.

I've never seen a system fail on the 8th pass of Memtest86+ that doesn't also fail (quicker) in Prime95.  Such an example might exist, but they're exceedingly rare.

Personally I use 3D Mark 11, 3D Mark Vantage, Unigine Heaven and a couple of games (like Crysis 2 with hires texture pack and DX11 Patch).

I always run Furmark too just to see how it holds up but I don't swear by it like some others. I don't see Furmark as the Prime95 for GPU's, it doesn't test a GPU like Prime tests a CPU. What Furmark does do well is generate as much heat as possible and in that respect I use it. 3D Mark and good game is a much better test of a graphics card IMO.

Regarding Furmark...  I don't know.  I guess I'm with you on that.  I have it installed on my test drive, and I've used it here and there to beat on my GPU temps to see if they're going to cause long-term problems, but... It just isn't testing in any kind of real-world usage scenario in any way, and GPUs don't work the same way as CPUs with heat and power consumption.  I basically agree with AMD and Nvidia.  Furmark isn't really very realistic, and it can cause otherwise-good GPUs to exceed specs (PCIe power specs mostly).

Otherwise, I totally agree.  A good 3DMark run and a game benchmark or two are the way to go when testing graphics performance.

Yeh .. I guess ... hesitatingly. I'm pulling difficult faces here. The OP asked for stress testing methodologodolyody, and that's what I gave. A quick sanity check, although useful and serving a purpose in its own right IF that's what is wanted, is not a stress test.

But let me re-emphasize, 2-3 hours is NOT a stress test, it doesn't 'prove' anything other than that there isn't a major issue. In my experience, if Prime^[1] fails in 2-3 hours you are already having issues, application crashes, memory reference errors or even STOP errors.

Yes, completely absolutely agreed.  I wasn't completely clear in my little note, but all I meant is that when you are NOT testing for final stability, but are actively working on the settings, it can be handy to run Prime95 for a single pass or so quickly, and then reboot and make additional changes.  Otherwise, overclocking would take forever, or (more likely) you just wouldn't test any interim steps.

The best thing to do is to test things one at a time, a little as you go.  But waiting 12 hours between each incremental BIOS tweak isn't practical.  So, a single-pass Prime run can be handy as a sanity check (or a quick OCCT run).  A lot of people "out there" recommend things like SuperPI (and it's multitude of relatives), but I've never found that to be a reliable quick-check tool.  By the time SuperPI shows errors, usually the system is so unstable that it is going to bluescreen if you breathe on it wrong.  SuperPI was useful way back in the day when we were testing Pentium II CPUs and single-core Athlons, but the machines are just so fast now that it isn't a reliable test.  You need to push on them till they heat up, and in a variety of ways to check all the different execution units.  SuperPI just doesn't do that.

Prime95 is the way to go.  That's all I meant.  When you are done futzing, you must do a full run to confirm final stability.

[glynor]

Mine right now scales from 1.05V (at idle, clocked at 1.6GHz) to 1.35V (load, clocked 4.4GHz).
Even at load it slightly fluctuates between 1.344 and 1.352, as the power regulators adapt.

Yep.  Digital VRMs and CPU integrated power management has changed everything.

Mine fluctuate too on my modern systems, even at load by similar amounts to what Nev is reporting.

[InflatableMouse]

It has nothing really to do with the PSU, it just works that way.
The CPUs power regulators will lower the voltage as the CPU is bored.

Mine right now scales from 1.05V (at idle, clocked at 1.6GHz) to 1.35V (load, clocked 4.4GHz).
Even at load it slightly fluctuates between 1.344 and 1.352, as the power regulators adapt.

The times of one fixed voltage are over. I'm sure i could configure my BIOS to behave that way, but there is no advantage in it.

Yeh, you're right. I'm still thinking about the old stuff sometimes, I'll edit my post to correct that.

[InflatableMouse]

Just a question there Nev, do you have your vcore set to AUTO, or did you select a specific number?

I thought I had done more to stabilize mine but apparently that is incorrect (must be old memories too lol), I just tested it. My vcore only fluctuates when I set it to auto. I have it set to 1.335v for 4.6GHz, that won't run on AUTO.

[Sandy B Ridge]

Wow, thanks for all the replies/tips.

SBR