• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Stress Testing for Beginners: Which tool should I use?

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

trents

Senior Member
Joined
Dec 27, 2008
AMDOverdrive? Prime95? OCCT? IntelBurnTest? Intel Extreme Tuning Utility? AIDA64 Extreme? Realbench?

So many choices! And this is not even an exhaustive list!

What are the weaknesses and strengths of each one? Having used all of the above at one time or another, I have some input for you that may be helpful.

Let me say first of all that in the end the goal is to achieve the max overclock that is truly stable on the lowest possible voltages with regard to how you use your computer. If you just want a snappier computer to use for undemanding tasks such as typing, email and internet browsing then your stress testing does not need to be quite as stringent as does someone who is playing demanding 3D games, folding, video rendering or crunching huge databases and others tasks that intensely work the CPU and memory. Casual users need not crank up the volts and the the heat just to be able to pass the most demanding stress tests. It would be a waste of time, electricity and probably also shorten the life span of your computer to some extent. But neither do you want to take stress testing shortcuts that may result in blue screens right in the middle of composing a long document or electronic filing of your taxes.

So I'm going to create three categories for dealing with stress testing tools: Heaters, Realistic and Pretenders.

Heaters
The most popular are Prime95, OCCT and IntelBurnTest. These stress testers put your CPU and memory to work doing a series of mathematical calculations, one after the other. Some of the calculations are more demanding than others and so you will see core/package temps rise and fall during the stress testing in a more or less rhythmic fashion. Each round of calculations is called an "iteration." Some of the iterations are much more demanding than others and will cause the highest temp spikes.

Prime95 - Historically, the most popular stress tester among overclockers. The defacto proof of stability has become passing 2 hr. of Prime95 blend mode, though some insist on a longer time and some prefer the small FFT test mode, which puts more stress on the CPU but less on the system RAM than the blend mode. Prime95 has various custom settings but one thing it lacks is a setting that would allow you to specify the time length of the test. This would be handy in the situation where you could not stick around to attend the process. It will grind away until you manually stop it. Prime95 is a real heater and longer tests produce temp spikes at intervals that are considerably higher than the temps seen during most of the testing cycle. Edit: As of v. 27.7 Prime95 utilizes AVX instruction sets. If you want an older version without AVX then download v.26.6. More recent versions of Prime95 utilize AVXII instruction that are accentuate this pattern. Because of this some people recommend using an earlier version of the program and one major motherboard manufacturer is on record as saying they no longer use it in their testing because it can drive temps up dangerously high. Pretty well maintained and can be used on both Intel AMD CPUs.

IntelBurnTest - Also a real heater. If you have plenty of RAM you can set the difficulty of the stress test at either Standard, High, Very High, or Extreme. If your amount of RAM is limited you will not be able to choose the higher difficulty levels. IntelBurnTest makes heavy use of RAM in it's testing method. Another nice feature is the ability to set the number of iterations employed during the test. Default is 10 but if you want the test to be harder and longer you can change that number to anything you like. Although not exactly a timer this feature allows you to put some rough limits on the length of the test for unattended use.The thing I like most about IBT is that, unlike Prime95, it heats up your cores to near max temp almost right away so that you have a pretty good idea of what your max temps will be during a longer, unattended test.The biggest drawback of IBT is that it seems to be an orphaned project. The last version was turned out in 2012, I believe. So it will not address some of the latest AVX instruction sets Edit: Apparently, version 2.54 of Intel Burn Test does use AVX but not AVX 2. It responds to my AVX offset in bios. As far as I know, it is the only version still available for download. IBT Can be used with both Intel and AMD systems.

OCCT - Edited: Allows you to set the time of the test and has nice built-in graphs for monitoring temps and voltages. Also has a "linpack" stress test option which itself has an AVX option. The intensity of the default stress test is much like Prime95 and so are the max temps produced. You can choose small, medium or large FFT iterations. Edit: Has recently been updated in the spring of 2018 to v. 4.5.1. Engages AVX (and I think, AVX 2) instructions. I really like this one. The intensity of it's stressing seems to be about the same as Prime95 but it has extra features not found in Prime95.

The heaters may be overkill for many but are certainly good for testing the limits of your cooling system.

Realistic
AIDA64 Extreme - Jack of all trades system info software, it has a built in benchmarking suite and, for the present purposes, a very good built-in stess tester with temp monitoring graph. Recommended as both a safe and effective stress tester by a major motherboard company and is gaining popularity as the defacto gold standard for stress testing. In the default mode, seems to stress in a more natural way (like real world applications do) than the heaters mentioned above and does not drive core temps so high.This may be a significant factor for a lot of folks who are using air coolers or modest water systems but who want to do a good job of testing for stability of their overclock. This software is free to try for 30 days and can be used on either AMD or Intel CPUs. It is updated frequently.

Intel Extreme Tuning Utility - Great software overclocking tool for Intel CPUs and comes with a decent but not great stress tester. Not quite as demanding as as AIDA64 Extreme and produces max temps several degrees lower. Stress tests focuses on either the CPU or the RAM but not both together. AIDA64 Ex tests both at the same time. But XTU is truly freeware and has the nice feature of being able to set the amount of time you want to run the test. For comparison sake, I could pass several hours of the XTU stress test at 5 ghz and a vcore of 1.285 but needed 1.305 vcore to pass hours of AIDA64 Extreme. Having said that, I ran the system for months at 5 ghz/1.285 vcore without any instability. I'm not a gamer so if I had been using the computer for more demanding tasks I might have well encountered instability.

Realbench
A newcomer by Asus that is truly freeware and will work on both Intel and AMD platforms. I have only played a little with it but like what I see. You can set the length of the stress test for anywhere from 15 minutes to 8 hr. and you can customize the amount of RAM used starting with a minimum of 4gb. It also has the option of stressing the GPU but I haven't experimented with that. The CPU stress test employs real world tasks like video encoding and file compression. Max temps generated seem to indicate the stress testing is fairly rigorous but the max temps are not extreme. In that sense it reminds me a lot of AIDA64 Extreme and XTU. Realbench, as the name indicates, also has real world app benchmarking tools. I see more and more overclockers migrating to Realbench. Edit: As of v.2.56 Realbench does use AVX but not AVX 2. Realbench requires the video card to have OpenCl ability. There is a problem with running the Realbench stress test on some systems such that it throws an error message to the effect of "OpenCL.dll missing". Updating your video driver to the latest version from the GPU company's website sometimes fixes this.

Pretenders
AMDOverdrive - For AMD systems only. AMDOD is multipurpose software with tools for overclocking, system monitoring and stress testing. Presently, is probably the only software that accurately reports the package temps of APU processors. The stress testing component is not very robust, however, and I do not consider passing it a reliable indicator that an overclocked system is stable. The stress tester timer can be set for up to 7 days. I use this program mostly to get the true package temp for APUs so that I can then factor in a temp differential when using HWMonitor, my monitoring utility of choice. I then do a real stress test with more robust testing tools.

Edit:What's AVX about? AVX stands for "Advanced Vector (E)Xtensions". If a CPU has AVX instruction sets it can process certain kinds of calculations coded for this technology very efficiently. As of now, AVX capable applications are not common, being found mostly in specialty software used by the scientific community. Who knows if AVX will find more common use in the future. AVX calculations cause CPUs to use a lot more power and produce a lot more heat. Recent vintage motherboards have an AVX offset option that causes the CPU core frequency multiplier to downclock when AVX code is encountered in an application. This can prevent overheating of the CPU and can also allow you to get a little higher overclock for running non AVX software. On the Intel side, you encounter AVX capable processors once you move up the ladder to the "i" series. Pentiums and Celerons do not have it. On the AMD side, we see AVX instructions first appearing in the "Bulldozer" generation of processors.

What does failing a stess test look like? BSOD (blue screen of death, computer lock-up, one or more "core workers" dropping out, or a spontaneous restart of the computer during testing means your overclock is not yet stable enough to pass the test. Some programs like IntelBurnTest may throw up a message saying your system was not found to be stable.

Can I hurt my CPU with stress testing? I'ts possible but not likely if you overclock in small increments. Most modern CPUs have thermal shutdown or throttle down mechanisms to protect themselves against this sort of thing. Make sure you monitor temps whenever stress testing. Keep max temps for current Intel CPUs under 90c and current AMDs under 65c. Motherboards are more likely to be damaged by overtaxing than are CPUs, especially over time, so watch those socket and VRM (Voltage Regulation Module) temps and know what the TDP (Total Power Draw) watt rating of your motherboard is. Many cheap motherboards have a TDP of around 95W and many CPUs at stock frequencies and voltages draw close to that or even more. These boards are not good candidates for overclocking some of the more power hungry CPUs. You are risking damage to the board if you try.

How long should I run a stress test to confirm stability? You will get a range of answers to this question from veteran overclockers. It depends on the stress test and it depends on how you use your computer. Personally, I find that two hours of Prime95 (blend option), IBT (using the Very High option) and OCC (medium FFT) ensure stability for my purposes. With AIDA64 Extreme and XTU I find that 5+ hr. gives me about the same level of assurance. As I already mentioned, AMDOD does not give me much assurance at all that my system is stable so I never use it alone. However you go about it, attend the process long enough to get a feel for what the max package/core temps will be as not to damage your system. Some of the more vigorous stress tests in short runs of 10-20 minutes may be good for saving time as you zero in on your final overclock but need to be followed up with something longer to provide confirmation. If the stressing software itself has no built in temp monitoring then find a stand alone one that will serve that purpose. The favorite around here seems to be the non pro version of HWMonitor.

Finally, I feel it needs to be said that perhaps the best approach to stress testing is to use more than one tool, as in getting a second and third opinion. And sometimes coming at stress testing from a different angle has benefit. For example, when testing overclock stability I like to include 3DMark which is primarily a GPU benchmarking tool. But it has several "physics" testing modules that stress the CPU and memory strongly. I have found it will sometimes reveal latent instability despite passing the stress tests of other tools. Also, if you want to push open the overclock envelope to the max, there is value in using the bios AVX offset so as to get the highest stable overclock when running non AVX applications. An offset of 2 is commonly used.

Updated 5/28/18
 
Last edited:

Kenrou

Member
Joined
Aug 14, 2014
Question on OCCT, seemingly hasn't been updated in year and a half (03 September 2014) is it still feasible ?
 

RollingThunder

Destroyer of Trolls & Spammers
Joined
Jan 7, 2005
Trents,

You might want to integrate Orthos with Prime95. Orthos IS Prime95 but with a much nicer GUI on top.
 
Last edited:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Nice little write up! :)

A couple things:

P95
- It was last updated 8/2015.
- AVX2 instructions I believe is what you want, not AVII?

OCCT
- What does stridency mean? Not sure it fits there?
- Also, it hasn't been updated in over 1.5 years (9/2014). Not sure I would call that up to date when comparing it to P95 (8/2015) and AIDA64 (weekly it seems, LOL).

AIDA64
- You can make it run like P95, check off FPU only and see. :)



Anyway, these days, I only use AIDA64. I leave it on the regular test and go. For my uses, typically gaming, some photo/video editing, and web/email etc, it hasn't steered me wrong. Typically I let it run for around 2-4 hours on the final shot. I RARELY bother to use two different tests. I didn't when P95 was my primary (ran Small FFT then Blend), and do not with AIDA64.
 
Last edited:

mackerel

Member
Joined
Mar 7, 2008
Prime95 is not linpack based. The code involves optimised assembler implementing FFTs, which are used as a relatively quick way to multiply massive numbers. It gets updated to take advantage of new instructions that help performance when available. This makes heavy use of floating point instructions, such as AVX and FMA (which came with AVX 2) on Intel CPUs. There is now speculation what AVX 512 might bring, if it ever reaches the masses.

What it tests will depend on the settings used. The ram required for a FFT is 8x the FFT size, so for example, 128k FFTs would take 1MB each. If the FFT fits inside the processor cache, it will not usefully test ram and will maximise stress on the FPU. FFTs above the processor cache size (lets say OVER 256k FFT for 2MB/core) will have to access ram. thus the memory controller will be stressed also, as well as some ram. The FPU load is reduced since it will typically be bottlenecked by the ram. There is a final setting on how it uses ram, in place or not. Normally the code will read/write to the same locations so it doesn't stress the ram much. So by setting an option accordingly, it forces it to write to different locations and thus test different ram locations also. I would not consider this a ram specific test in itself. For indication, real world Prime95 work is currently running 4096k FFT size, so well into ram performance dependant.

As for it being "too stressful" that comes down to how you define stability. If you never push a computer to a certain workload, does it need to be stable in that workload? The math library of Prime95 is available for general use, and other software can also take advantage of it. I use some of that other software, so my modern systems are effectively running Prime95 like loads continuously as a matter of routine. To me saying you shouldn't run it as it is too stressful simply means whatever you're not running it on is under-designed. This might be fair enough if it is a form factor constrained all-in-one or laptop for example, but inexcusable in a desktop class system.

I'm not so familiar with its performance on AMD CPUs since they're lagging significantly in performance in the instruction types used. It is not that instructions aren't available, but that they don't give the performance increase when used like on Intel.

When I looked up IntelBurnTest, it seems too old to use FMA (AVX 2) instructions so testing will likely be using AVX. I'm not sure what is mean by the claim that Prime95 needs to warm up, unless you're using the blend mode where it changes FFT sizes thus tests different things at different times.

Intel Extreme Tuning Utility has the alternative name Intel XTU at least where I see it most. I haven't come across the use of IXT before.
 

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
Prime95 is not linpack based. The code involves optimised assembler implementing FFTs, which are used as a relatively quick way to multiply massive numbers. It gets updated to take advantage of new instructions that help performance when available. This makes heavy use of floating point instructions, such as AVX and FMA (which came with AVX 2) on Intel CPUs. There is now speculation what AVX 512 might bring, if it ever reaches the masses.

What it tests will depend on the settings used. The ram required for a FFT is 8x the FFT size, so for example, 128k FFTs would take 1MB each. If the FFT fits inside the processor cache, it will not usefully test ram and will maximise stress on the FPU. FFTs above the processor cache size (lets say OVER 256k FFT for 2MB/core) will have to access ram. thus the memory controller will be stressed also, as well as some ram. The FPU load is reduced since it will typically be bottlenecked by the ram. There is a final setting on how it uses ram, in place or not. Normally the code will read/write to the same locations so it doesn't stress the ram much. So by setting an option accordingly, it forces it to write to different locations and thus test different ram locations also. I would not consider this a ram specific test in itself. For indication, real world Prime95 work is currently running 4096k FFT size, so well into ram performance dependant.

As for it being "too stressful" that comes down to how you define stability. If you never push a computer to a certain workload, does it need to be stable in that workload? The math library of Prime95 is available for general use, and other software can also take advantage of it. I use some of that other software, so my modern systems are effectively running Prime95 like loads continuously as a matter of routine. To me saying you shouldn't run it as it is too stressful simply means whatever you're not running it on is under-designed. This might be fair enough if it is a form factor constrained all-in-one or laptop for example, but inexcusable in a desktop class system.

I'm not so familiar with its performance on AMD CPUs since they're lagging significantly in performance in the instruction types used. It is not that instructions aren't available, but that they don't give the performance increase when used like on Intel.

When I looked up IntelBurnTest, it seems too old to use FMA (AVX 2) instructions so testing will likely be using AVX. I'm not sure what is mean by the claim that Prime95 needs to warm up, unless you're using the blend mode where it changes FFT sizes thus tests different things at different times.

Intel Extreme Tuning Utility has the alternative name Intel XTU at least where I see it most. I haven't come across the use of IXT before.
:thup:
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
Good job Trents, nice to have a bit of a guide. Thanks for the time.

When I looked up IntelBurnTest, it seems too old to use FMA (AVX 2) instructions so testing will likely be using AVX. I'm not sure what is mean by the claim that Prime95 needs to warm up, unless you're using the blend mode where it changes FFT sizes thus tests different things at different times.

There is an Intel burn test available that does use AVX2, I use it for quick testing on AMD setups. Typically people have trouble with it on Intel.

View attachment IBT AVX.zip
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Trents,

You might want to integrate Orthos with Prime95. Orthos IS Prime95 but with a much nicer GUI on top.

Thanks for the suggestion. Haven't used it but I will check it out and likely add that. Is it otherwise indentical to Prime95?

- - - Updated - - -

Good job Trents, nice to have a bit of a guide. Thanks for the time.



There is an Intel burn test available that does use AVX2, I use it for quick testing on AMD setups. Typically people have trouble with it on Intel.

View attachment 177859

Yes, I have heard that but never have been able to find it. Johan45, could you send me a URL in addition to the download link? Is this also a Meresene.org project or something independent of that group?

Prime95 is not linpack based. Thanks for correcting me on that. The code involves optimised assembler implementing FFTs, which are used as a relatively quick way to multiply massive numbers. It gets updated to take advantage of new instructions that help performance when available. This makes heavy use of floating point instructions, such as AVX and FMA (which came with AVX 2) on Intel CPUs. There is now speculation what AVX 512 might bring, if it ever reaches the masses.

What it tests will depend on the settings used. The ram required for a FFT is 8x the FFT size, so for example, 128k FFTs would take 1MB each. If the FFT fits inside the processor cache, it will not usefully test ram and will maximise stress on the FPU. FFTs above the processor cache size (lets say OVER 256k FFT for 2MB/core) will have to access ram. thus the memory controller will be stressed also, as well as some ram. The FPU load is reduced since it will typically be bottlenecked by the ram. There is a final setting on how it uses ram, in place or not. Normally the code will read/write to the same locations so it doesn't stress the ram much. So by setting an option accordingly, it forces it to write to different locations and thus test different ram locations also. I would not consider this a ram specific test in itself. For indication, real world Prime95 work[/U] is currently running 4096k FFT size, so well into ram performance dependant.

As for it being "too stressful" that comes down to how you define stability. If you never push a computer to a certain workload, does it need to be stable in that workload? The math library of Prime95 is available for general use, and other software can also take advantage of it. I use some of that other software, so my modern systems are effectively running Prime95 like loads continuously as a matter of routine. To me saying you shouldn't run it as it is too stressful simply means whatever you're not running it on is under-designed. Here, I think you are saying what I already said in my third paragraph. Please go back and read it. This might be fair enough if it is a form factor constrained all-in-one or laptop for example, but inexcusable in a desktop class system.

I'm not so familiar with its performance on AMD CPUs since they're lagging significantly in performance in the instruction types used. It is not that instructions aren't available, but that they don't give the performance increase when used like on Intel.

When I looked up IntelBurnTest, it seems too old to use FMA (AVX 2) instructions so testing will likely be using AVX. I'm not sure what is mean by the claim that Prime95 needs to warm up, unless you're using the blend mode where it changes FFT sizes thus tests different things at different times. Yes, I am referring to the blend mode which I think is most commonly used.

Intel Extreme Tuning Utility has the alternative name Intel XTU at least where I see it most. I haven't come across the use of IXT before. Thanks, I'll fix that.
 
Last edited by a moderator:
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Nice little write up! :)

A couple things:

P95
- It was last updated 8/2015. Fixed that. Thanks.
- AVX2 instructions I believe is what you want, not AVII? Fixed that. Thanks.

OCCT
- What does stridency mean? Not sure it fits there? Changed "stridency" to "intensity". Thanks.
- Also, it hasn't been updated in over 1.5 years (9/2014). Not sure I would call that up to date when comparing it to P95 (8/2015) and AIDA64 (weekly it seems, LOL).

AIDA64
- You can make it run like P95, check off FPU only and see. :) Will add a line about being customizable.



Anyway, these days, I only use AIDA64. I leave it on the regular test and go. For my uses, typically gaming, some photo/video editing, and web/email etc, it hasn't steered me wrong. Typically I let it run for around 2-4 hours on the final shot. I RARELY bother to use two different tests. I didn't when P95 was my primary (ran Small FFT then Blend), and do not with AIDA64.
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
That file I left was taken from OCN piledriver thread. IIRC it's an altered version of the original IBT test and I think it was done by the original author( as you see rev level is the same but the test isn't). Because of the addition of the AVX it wasn't popular for Intel with all the hype about killing your haswells with P95.
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Trents,

You might want to integrate Orthos with Prime95. Orthos IS Prime95 but with a much nicer GUI on top.

RT, I cannot find a download link for a current version of Orthos. What I have found is either ancient or makes use of one of those malware laden downloader utilities.

- - - Updated - - -

Thanks, guys. I have implemented most of your suggestions.

- - - Updated - - -

That file I left was taken from OCN piledriver thread. IIRC it's an altered version of the original IBT test and I think it was done by the original author( as you see rev level is the same but the test isn't). Because of the addition of the AVX it wasn't popular for Intel with all the hype about killing your haswells with P95.

Then I think I'll leave well enough alone.

Deleted all the stuff about linkpack. It is information not needed by beginners anyway.
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
Trents,

This is not critical, I have migrated AIDA64, but others may not have and like Prime95.

Looks like a safe download from here if you wish:

http://www.softpedia.com/get/System/Benchmarks/Orthos-Prime.shtml

Yeah, that's where I got it from earlier this morning. But from looking at the "About" info it's dated 2004!

Nice little write up! :)

A couple things:

P95
- It was last updated 8/2015.
- AVX2 instructions I believe is what you want, not AVII?

OCCT
- What does stridency mean? Not sure it fits there?
- Also, it hasn't been updated in over 1.5 years (9/2014). Not sure I would call that up to date when comparing it to P95 (8/2015) and AIDA64 (weekly it seems, LOL).

AIDA64
- You can make it run like P95, check off FPU only and see. :)



Anyway, these days, I only use AIDA64. I leave it on the regular test and go. For my uses, typically gaming, some photo/video editing, and web/email etc, it hasn't steered me wrong. Typically I let it run for around 2-4 hours on the final shot. I RARELY bother to use two different tests. I didn't when P95 was my primary (ran Small FFT then Blend), and do not with AIDA64.

My experience is that when my overclock isn't quite stable AIDA64 will often crash at about the five hour mark.
 
Last edited by a moderator:

EarthDog

Gulper Nozzle Co-Owner
Joined
Dec 15, 2008
Location
Buckeyes!
I don't run it that long... ever... well, intentionally anyway, LOL!

That's the thing with stress testing, there isn't an answer on the best or what is good for me... etc.
 
OP
trents

trents

Senior Member
Joined
Dec 27, 2008
I don't run it that long... ever... well, intentionally anyway, LOL!

That's the thing with stress testing, there isn't an answer on the best or what is good for me... etc.

That is certainly true!
 

Johan45

Benching Team Leader Super Moderator
Joined
Dec 19, 2012
I think I undestand the IBT thing a bit better now and why it may not work on Intel very well it's a patch for AMD so it will use AVX did some reading here and here
 

wingman99

Member
Joined
Dec 10, 2003
I do like running Prime 95 on blend so it moves a lot of data in and out of the CPU cache and stresses the CPU without much heat.