Stress Testing for Beginners: Which tool should I use?

[trents]

I guess it is(?), but, to me, its not remotely common. First I heard of it honestly. I wouldn't worry about it and stick to the well known ones.

Some people who use Realtemp as a temp monitor may be using it's load tester as well and may not be aware that it doesn't really do a good job. The other thing is, AMDOD was there all by it's lonesome self in the "Pretenders" category.

 

[EarthDog]

Something I dug up: http://www.techpowerup.com/forums/threads/realtemp-general-discussion.64185/page-16#post-1550447

Load Tester is the direct opposite of Prime95. LoadTester puts a very gentle load on your CPU and is good for testing turbo boost on the new Core i5 / i7 processors as well as EIST testing.

I'd drop it personally since it isn't really a stress test in the first place.

And here I was thinking it used P95 (that is what the sensor test uses I believe)!

 

[trents]

I wouldn't call Realbench good stability test. On some platforms it's randomly crashing during some test stages. It's probably because for example OpenCL stage requires good and stable drivers. On the other hand if it runs 100% stable on stock then it's quite good tool to check stability of overclocked CPU.

XTU is running Prime95 process. You can check it in task manager when you run benchmark. (Didn't see anything that looked like a Prime95 process in Task Manager. I did see linpack xeon x64 in there when running the XTU stress test.) Also for some reason benchmark is more demanding and is crashing faster than stability test.

Prime95 is good stability test but is not perfect. The same as most other programs. Best is to run couple of applications as all are using different settings and computer resources. I'm not sure if anything more than ~2-3h in Prime95 has any point. It's better to set 3DMark11/Vantage/2013 in looping on CPU/Physics and Combined tests for next 2h.

AIDA64 is good stability test but you have to know how to use it. As ED said if you run only 2 first tests then CPU load will be higher than when you run all tests at once. If you test memory stability then always run memory and cache together as it loads memory controller. Btw. AIDA64 is testing 95-99%+ free memory. There is always couple of MB left so OS/applications won't freeze.

I just run AIDA64 in default mode. The cache and memory boxes are checked by default.

 

[ShrimpBrime]

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 runs for 60 minutes and that cannot be customized. 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.

Hmm. Dunno bout this. Mine shows a slider that goes to seven days. I recall that being there for many years now.

See? Heats the ol FX processors quite well. Some one oughta' try it some time.

 

[trents]

I'll have to check that out.

Update: You are correct. There is a slider for setting the timer for up to 7 days. Never noticed that before. I just assumed it was a progress bar. I will amend my statement in that section of the first post.

Thanks.

 

[Woomack]

In older Windows you could use calculator to force full cpu load. Any high number + n! was making cpu to count for a long time at 100% load. I see it's not working in Win10 and I get overflow error. It was stressing CPU to the point you couldn't even move mouse.

About that Prime95 process in XTU, I'm sure it was in older XTU versions but maybe it was changed. Now I don't have any PC with CPU that could run XTU. If I won't forget then will check it after work.

 

[Johan45]

I was just going to mention the slider. Using it right now for some testing. I do wonder if there is a way to make it more stressful? I have never used AMD OD as an actual stress test before, just overclocking. So my knowledge on it is kind of limited to my experience. I do wonder if it is enough for general/gaming purposes?

 

[Kenrou]

Realbench got a nice tweak seemingly, check it out : "The Handbrake h.264 test in Realbench is there to test a specific type of loading -- near 100% loading on all threads/cores, heavy AVX use, plus heavy synchronized cache and memory I/O. It's a torture test to the redline on everything. The other tests allow more asynchronous operation. The Heavy Multitasking test includes Handbrake h.264 encoding, but it's a subset of other tasks (it's not permitted to take over like the singular h.264 Handbrake test) which will mask some stability problems that the singular h.264 test will reveal."

That explains why it drove my 4.7ghz 1.38v to the low 80s

 

[wingman99]

Realbench got a nice tweak seemingly, check it out : "The Handbrake h.264 test in Realbench is there to test a specific type of loading -- near 100% loading on all threads/cores, heavy AVX use, plus heavy synchronized cache and memory I/O. It's a torture test to the redline on everything. The other tests allow more asynchronous operation. The Heavy Multitasking test includes Handbrake h.264 encoding, but it's a subset of other tasks (it's not permitted to take over like the singular h.264 Handbrake test) which will mask some stability problems that the singular h.264 test will reveal."

That explains why it drove my 4.7ghz 1.38v to the low 80s

How does it compare to Prime95 for temperature?

 

[Kenrou]

On my setup roughly the same as Prime95 blend, had to increase voltage by +0.06v to become stable enough to pass the h264 coding, obviously YMMV ?

 

[Kenrou]

Out of curiosity, didn't see it mentioned anywhere so far (including other threads), does everyone use the standard blend test or are you supposed to increase the ram amount ?

 

[EarthDog]

Depends on the user. When I used P95, I left it alone... I don't what one is 'supposed' to do though. It would make sense to increase it to just below your actual amount but... leaving it at default has never steered me wrong before.

 

[wingman99]

I just use prime95 at default standard blend test with rounding error check. There is less heat with default settings.

 

[mackerel]

The descriptions on the Prime95 torture test window pretty much describe what happens.

Small FFTs = fits inside processor cache, stresses FPU the most, doesn't really test ram. If you get errors on this, look at the CPU.
Large FFTs = doesn't fit in processor cache, so memory interface and whatever ram it uses will also be tested. If you have no errors in small but fail large, look at memory system.
"In place" or not means if the data is kept and accessed in the same location (like normal software would), or if it is intentionally moved around (effect of testing more ram). Personally I wouldn't rely on Prime95 to find problems in ram itself, but it could show weaknesses if the CPU ram interface wasn't stable.

If you want to customise the FFT size, rough guide is FFT size multiplied by 8 is the number of MB required for the working data. e.g. 128k FFT would require 1MB. That assumption doesn't allow for code or other overheads so a bit more might be required.

 

[Kenrou]

OCCT updated, 27 April 2016 :

http://www.ocbase.com/index.php/45-occt-4-4-2-available

 

[danialalhabshi]

I know this is last year's thread but excellent write up! Nice summary of each and very helpful to beginner overclockers (myself included). I just finished building my first gaming PC in a while. See my configuration below.

After long days of research and look around I decided to overclock my i5 7600K. My idle CPU temp is between 28-31C, and moderate load temp ~50-52C. I use CoreTemp and CPU-Z as my monitoring tools. Now, I know there are many theories on why you should keep your Vcore static but after running a couple hours on AIDA64 (averaging 66C with max 76C @1.240V-1.285V), Intel Extreme Tuning Utility or IXTU (averaging 62C with max 67C @1.230V-1.265V) and RealBench (averaging 73C with max 84C @1.230V-1.250V) separately each on adaptive voltage all @4.6GHz I find that's where I like to rest my OC for now. I use the PC mainly for moderate gaming and some multitasking jobs here and there. I rarely do any video rendering or heavy editing so I don't think I really need it over that 4.6 mark (yet). My question is, with the voltage, if it ranges from 1.230V-1.285V between the three tools, should I really be giving my CPU that extra voltage it didn't need and make it static at say, 1.285V (max of them all) or do I let my CPU enjoy the lower voltage when it desires?

And another thing, before using any of the other stress tools, I started my OC tests on prime95 (v28.10). Funny thing was, I could barely make it 20 secs @4.5GHz even after I crank the voltage up to 1.325V. And each time it 'fails', my PC didn't BSOD'ed or froze, it ran completely fine but it was always just that one core worker #4 coming up with the "FATAL ERROR: Rounding was 0.5, expected less than 0.4" thing. And staying true to its 'class', p95 brought the temp right up to 95-96C of which I stopped it right away and let the CPU rest for a few hours. With that high of a temperature at so low of a clock and not even that high of a voltage I refuse to believe my system was unstable so I looked up forums after forums and came across those other suggestions like AIDA64, IXTU, RB and so forth. And truly enough I got all those bracketed number said above @4.6GHz (again, AFTER that horrid test on p95) without any single problem. 1-2 hours each. And a few times over. I even left the IXTU monitoring page on while testing with AIDA64 and RB to make sure I definitely didn't have any thermal throttling. And my benchmark scores too with Cinebench R15, IXTU and RealBench are roughly in the top 10% of the whole world. So why did p95 fail me that easily still? Some say faulty RAMs usually cause that fail, but my RAMs passed all memory stress on the other stress tools.

I think it's fair to say my rig is stable to my needs at my current overclock, right? Anyways, thanks again for the excellent post. I find some assurance reading it after the mini-heart attack p95 gave me a few nights back.
I know this is more like a general thread for stress tools but knowing that you have used all of the tools I used (and we use the same chipset!), I thought it would be appropriate to direct those questions here. Thank you!


The rig:
CPU: Intel Core [email protected] ([email protected]) Un-delidded
Motherboard: GIGABYTE Z270 Gaming K3 ATX
CPU Cooler: Cryorig A40 Ultimate Hybrid Cooler
RAM: Corsair Vangeance LPX 16GB (2x8GB) DDR4 DRAM 3000MHz
GPU: ASUS GeForce GTX 1070 8GB Dual Series O8G
Monitors: 27" ASUS Gaming LED 1440p & 23" LG Flatron LCD 1080p
Case: Phanteks Eclipse P400S Tempered Glass Glacier White Mid ATX
Fans config: Front mounted 240mm radiator intake with push configuration, 1x 120mm rear intake, 1x 120mm top exhaust
PSU: EVGA Supernova G1 650W
Storage: Samsung 250GB 850 EVO SSD & WD Blue 1TB HDD
OS: Windows 10 Home 64-bit

 

[mackerel]

So why did p95 fail me that easily still? Some say faulty RAMs usually cause that fail, but my RAMs passed all memory stress on the other stress tools.

If you want a stress test, you get one with Prime95. It is doing a defined mathematical task, and if results are not as expected, there is probably a problem with the hardware. The routine does multiplication of two large integer numbers using floating point maths. Why do this is rather complicated, but it is the most compute efficient way we have right now. Of course, floats are not integers, so they have to be checked and converted as necessary. This is where the rounding error comes in. The code is designed so that the rounding error should always be small enough that it wont give a false result. In practice, I've found stability will decrease with temperature. At a set voltage and clock, I can be stable at 70C, and unstable at 80C. A hot day, or dust buildup can tip the balance.

The traditional argument against Prime95 is that it doesn't represent other more typical uses. You can choose to be Prime95 unstable if other things appear stable, but you are still Prime95 unstable.

There are rare occasions where errors have been found in Prime95, most recently when the math library was used in other software under some situations with multi-threads enabled. Where verified as a bug they are generally squashed in due course. Unless you're doing something rather new and uncommon with it, the odds are it is the hardware not the software.

 

[danialalhabshi]

In practice, I've found stability will decrease with temperature. At a set voltage and clock, I can be stable at 70C, and unstable at 80C. A hot day, or dust buildup can tip the balance.

The traditional argument against Prime95 is that it doesn't represent other more typical uses. You can choose to be Prime95 unstable if other things appear stable, but you are still Prime95 unstable.

I can see you point there. And I do believe your own system's 'stability' is also sort of heavily dependent on your usage. At the moment I really am enjoying the extra in-game FPS and task transitions speed with that extra clock eventhough I know I'm Prime95 unstable. And also for now I know my temps are always under 65C at moderate-heavy load and my voltage is well under 1.285V at any given time so I'm not too overly concerned about frying the CPU anytime soon (it is however, winter here at the moment averaging 14-19C room temp). I keep my Core Temp always on top on my second screen so I always know what my temps are doing. If I continue to run my PC like this, knowing that I will never go full on 100% at any time other than stress testing, will my CPU still suffer from running with that one probable hardware issue found in p95? Still though, knowing that one core worker not passing the p95 test at the back of your head does bug you at times.

There are rare occasions where errors have been found in Prime95, most recently when the math library was used in other software under some situations with multi-threads enabled. Where verified as a bug they are generally squashed in due course. Unless you're doing something rather new and uncommon with it, the odds are it is the hardware not the software.

So if I couldn't find a stable voltage to stabilise that desired clock on p95 after pushing it up to say, 1.38V (which I haven't), would you suggest I go down the path of narrowing down which hardware was faulty? It is rather odd given that I got all the parts delivered to me only a few weeks ago (yes I do know even the newest of hardwares can have faults at stock). But, if a certain said hardware really is faulty, wouldn't you expect it to also be unstable at a much lower clock for example? In my case I passed the p95 test at 4.4GHz but not 4.5 or higher.

 

[wingman99]

Test the memory overclock first with Pime95 then overclock the processor. My rig is prime95 v28.10 stable at 4.5GHz at 90c for 7 hours.

 

[mackerel]

I can see you point there. And I do believe your own system's 'stability' is also sort of heavily dependent on your usage. At the moment I really am enjoying the extra in-game FPS and task transitions speed with that extra clock eventhough I know I'm Prime95 unstable. And also for now I know my temps are always under 65C at moderate-heavy load and my voltage is well under 1.285V at any given time so I'm not too overly concerned about frying the CPU anytime soon (it is however, winter here at the moment averaging 14-19C room temp). I keep my Core Temp always on top on my second screen so I always know what my temps are doing. If I continue to run my PC like this, knowing that I will never go full on 100% at any time other than stress testing, will my CPU still suffer from running with that one probable hardware issue found in p95? Still though, knowing that one core worker not passing the p95 test at the back of your head does bug you at times.



So if I couldn't find a stable voltage to stabilise that desired clock on p95 after pushing it up to say, 1.38V (which I haven't), would you suggest I go down the path of narrowing down which hardware was faulty? It is rather odd given that I got all the parts delivered to me only a few weeks ago (yes I do know even the newest of hardwares can have faults at stock). But, if a certain said hardware really is faulty, wouldn't you expect it to also be unstable at a much lower clock for example? In my case I passed the p95 test at 4.4GHz but not 4.5 or higher.

Fault is probably too strong a word to use. One core erroring earlier than others is not unexpected. They are not all equal, and maybe it is the silicon, or maybe uneven temperatures. All 4 wont necessarily fall over at the same time. CPUs are sold at a given speed, and through overclocking we can tap into more of their potential. You're going to push too far at some point.

I would add, depending on the FFT size, the loads on the CPU will vary. I wrote about it earlier in the thread, but in short, small FFT is kept in CPU, large FFT will also hit the memory controller and ram too, with potentially less load on the cores. You could fail one and not the other.