• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Stability testing recommendations?

trickson

OH, I have such a headache
Joined
Dec 5, 2004
Messages
7,595 (1.02/day)
Location
Planet Earth.
System Name Ryzen TUF.
Processor AMD Ryzen7 3700X
Motherboard Asus TUF X570 Gaming Plus
Cooling Noctua
Memory Gskill RipJaws 3466MHz
Video Card(s) Asus TUF 1650 Super Clocked.
Storage CB 1T M.2 Drive.
Display(s) 73" Soney 4K.
Case Antech LanAir Pro.
Audio Device(s) Denon AVR-S750H
Power Supply Corsair TX750
Mouse Optical
Keyboard K120 Logitech
Software Windows 10 64 bit Home OEM
I have been out of touch for some time and I remember a time where we had benchmark programs that "WE all agreed upon as the official test to measure your system against others. I know the old ones but what are "WE" using today? Is there one free benchmark that is used and if so can some one please tell me.
I also remember a time when posting them along with a stable over clock was a MUST. I am not seeing this any more shy?
 
There's a pretty good selection now, I've probably missed loads but here's a quick list off the top of my head. There's threads discussing and keeping track of scores for most of these.

Graphics: There's still the 3Dmark benchmarks, Firestrike, timespy etc are the latest. Unigine does a few cool benchmarks, heaven, valley and superposition.

CPU wise there's Cinebench, Realbench, HWBot X265..

memory I tend to use the Aida64 memory & cache benchmark

Stability testing, there's still prime95, realbench again has a nice stability test, aida64, Memtest64 (by w1zzard!). Memtest86+.

That should keep you entertained for a while! I'm signing off for tonight, have a good evening trickson ;)
 
Thank You getting them now! Yeah this should keep me going. I need to test my OC and want to make sure it is stable.
 
I use OCCT primarily for CPU, GPU and system/PSU. Then some real world usage like games and encoding. Keep Memtest, Prime, Asus ROG Bench (encoding tests) and Aida64 around but don't generally need them.

I'm on the road or I'd post more. But I'd say try all the suggested options and see what works for you.
 
I use OCCT primarily for CPU, GPU and system/PSU. Then some real world usage like games and encoding. Keep Memtest, Prime, Asus ROG Bench (encoding tests) and Aida64 around but don't generally need them.

I'm on the road or I'd post more. But I'd say try all the suggested options and see what works for you.

RYZEN BLENDER!

There's a pretty good selection now, I've probably missed loads but here's a quick list off the top of my head. There's threads discussing and keeping track of scores for most of these.

Graphics: There's still the 3Dmark benchmarks, Firestrike, timespy etc are the latest. Unigine does a few cool benchmarks, heaven, valley and superposition.

CPU wise there's Cinebench, Realbench, HWBot X265..

memory I tend to use the Aida64 memory & cache benchmark

Stability testing, there's still prime95, realbench again has a nice stability test, aida64, Memtest64 (by w1zzard!). Memtest86+.

That should keep you entertained for a while! I'm signing off for tonight, have a good evening trickson ;)
 
I've given up on synthetic tests and whatnot. If my rig stays stable running Handbrake that's good enough for me. And it does.
 
Aida64 stability FPU for extreme temp stress testing, Aida64 Stability test CPU / FPU / Cache / Memory for total CPU stability.

For GPU Firestrike stability test.

Whole system test, Run Battlefield 4 or Battlefield 1 if somethings not stable CPU / GPU or memory wise it will crash to desktop A LOT!
 
For me..

CPU = Chess engine called Stockfish 9 x64 bmi2 1 hour is more than enough.
NB (Cache) & Memory = HCI memtest 100% is enough. Some may say 400-500%+ but after my try 100% is OK. I never encounter from BSOD with 100% pass.
GPU = Games only. Test all games that you played and if it didn't have problem then it is good.
 
I use ASUS realbech. At least goes for a 12h max RAM test.
 
On good game engines, an unstable system will have a fail in less than 5 mins. Witcher 2 or 3, BF4 or BF1 are ideal for this imho. Synthetic benchmarks sometimes are pushing the whole system but parts of it.
 
I see no value to synthetics except as follows:

a) I use prime 95 to cure the TIM with a couple of quick runs to bring up the CPU temp to 85C for a few seconds with minumal fan speeds and take it right down. Once cooled to stable temp, "rinse and repeat", As long as ya not using a paste that takes 200 hours to cure (i.e. AS5), you'll be done in a hour.

b) I use Furmark to adjust my radiator fan and pump speeds. On current box, the GPus were at 39C at 1200 rpm, and 44C at a point where a fans were inaudible.... 44C it is.

We stopped using synthetics when AVX and subsequent new instruction sets came into play. When overlcocking, most uses have the temperature and voltage limits that they want to use and AVX thru that all outat whack as voltages would pop up 0.13v and tempsas much as 10C when AVX was present. Synthetics present a load that after you dial in ya OC, the system will never see again. So what's the point.

1. Because it's an extreme load, its kinda like testing your SUV that you intend to pull your 400 pound boat / trailer to the beach 15 miles away buy towing a low-boy flatbed with a bulldozer up the Rocky Mountains. Not rally representative in most cases what that PC will be doing from Day 2 thru 1500

2. With regard to those temperature limits, synthetics will push voltage and temps higher than anything else you will ever do. So if your OC is P95 stable at 4.6 GHz and a constant 1.40 volts and highest core temps at 80C, when you run whatever you are going to run, that OC might only be 1.35 w/ 71 mac core temp under every day usage. In other words... lookes like ya might be able to hit 4.7 or 4.8 with more reasonable loads.

3. Finally, when first looking at alternatives to synthetics, we found that a 24 hour stable P95 OC would sometimes fail under RoG Real Bench multitasking test in < 2 hours.

So the approach these days is ....all rad and case fans on max speed

A. Cure the TIM, as above so you don't have temps changing throughout testing runs.

B. Run everything at stock settings, lower voltages incrementally till pass RoG RB for 2 hours and use this as baseline.

C. From here, take two approaches depending on time available....a) bump CPU multiplier by +1 over turbo and incrementally by 1 from there to get stable OC w/lowest voltages / temps possible. Turn on XMP and rerun... or b) do some reach and look for the average / medium CPU Multiplier OC and skip all the ones in between; work up or down by +1 depending on results. Save results for each CPU multiplier for each Multiplier

D. Once at that point, I'll do a 4 hour run of RoG RB as the CPU stability test. Then I'll move up cache multiplier to CPU Multiplier -3 . If just a gaming box, can leave at default but w/ graphics apps, there can be a drop off if more than 3 below ... going higher has not shown any benefit in my limited testing.

E. At that point, before bed time, memtest+ overnight (12 hours)

F. Finally GPU testing..... Find max stable core OC with Furmark (memory at stock); then find max stable memory OC with Furmark.

G. At that point, it's about finding the best combo between fastest memory and fastest core and the determination is not by those numbers. The you can get higher fps with lower core or memory speeds ... so i stop looking at those and average the GFX benchmark results using the "usual suspects" benchmarks.

4. Once done, I have never had an application or game that warranted any CPU or memory adjustments. On the GPU side, I have found it necessary to keep at least 3 profiles.

1. What I got from above testing.... possible adjusted downwards if get problems in a significant number of games.

2. A lowered setting because I was experiencing crashes at some point in a game. In Metro 2033 I wold experience this when coming up from underground and looking out over the junkyard ... would have to close game, drop to this setting, climb out from underground , then save and go back to high setting.

3. A still lowered setting for anything that yhad the word "Battlefield" in it :)
 
I just play whatever game I'm into at the time, that way I know if it will do the job or not.
 
I just play whatever game I'm into at the time, that way I know if it will do the job or not.

I should add that Ashes of the Singularity is one of the best for system stress testing IMHO. Do a skirmish with a bunch of AI, crank the graphics up, and play a round of pure destruction. It is the only game I play that loads my systems so heavily. Yet is scalable to play smoothly on my sons rigs in my sig. :)
 
aida and realbench for me.

both for stress testing and bench marking results.
 
https://www.techpowerup.com/download/lost-planet-2-benchmark/

Test 2.

If your comp passes/completes test 2 it's stable. I don't bother with all that other shit.
I dont like you.... I ran this on what as been my stable Ryzen 1600@3.9 - 1.32v and it crashed towards the end of the 2nd test :laugh: just to confirm I reset mem oc and re-ran the test and again it crashed, set CPU to auto and it didnt crash, I now need 1.335v at 3.9 to get it not to crash :p :toast:
 
Is it really that simple? Will try it anyway, and see if I can get rid of Realbench, OCCT, IBT, and Prime95.

Yep. It's now my/our not so secret secret now.
 
I'm basically going to rip this from what I posted in another thread:

If you want to test your CPU overclock with a few other stability tests I recommend the following:

Aida64 - You can just use the Extreme Trial Version, not the free edition. Under the Tools there is a System Stability test. I run the CPU, FPU, and Cache stress test for at least 12 hours before I consider the CPU Stable. The memory stress test is good too if you are overclocking your memory.

x264 Stability Test - It is best to run this test from a HDD you have in the system if possible, not a SSD, because it does write a lot to the disk. For a quick test I like to use a 5 loop run. But for my final stability test I set it to infinity and then let it run for at least 12 hours. Of course, quick is a relative term depending on your CPU. Because this program is actually doing video encodes, the amount of time it takes to complete a loop greatly depends on your CPU. So for dual-cores I tend to use 2 or 3 loops as a quick test instead of 5.

OCCT - This one has two CPU stability tests in it. One they developed(though I think it is pretty much Prime95 based) and a Linpack stress test. Again, I run both for 12 hours each.

IntelBurnTest - This is another Linpack stress test like the one in OCCT, so I don't really even use this anymore. I just use the one in OCCT. But it isn't a bad one to have if you want to use it.

I really only stability test with Aida64, x264 Stability Test, and OCCT anymore. IMO, that gives a very good coverage on CPU use cases. I don't use Prime95 anymore because the OCCT stress test seems to be extremely similar to it, so it would just be a redundant test. The same goes for IntelBurnTest, the Linpack test in OCCT does basically the same testing so I don't bother running both. Yes, I do run each test for 12 hours after I finalize my overclock, for a total of 2 days(more like 2 and a half actual due to gaps between running tests), but that is what I like to do to give me the best chance of a long lasting stable overclock. I don't like to mess around with finding out my overclock isn't actually stable a few months down the road when I decide I want to do something different that just gaming like encode video. So I just cover all the bases when I set up the overclock so I worry less. I know this much stability testing is not for everyone, and some will say any amount of stability testing is a waste and you should just play games and hope they don't crash, but that isn't what I like to do.
 
You could use 30 stress tests if ya like or you could just pick the few that address the goals for your PC. There's a link to a nice list below.

If it's just gaming box, I get these in all the time where ... the post it says "I tested with with Heaven and this and that and it's fine but it fails when I do [insert whatever here). Sometimes it's a different game, sometimes it's GFX program. Testing with synthetics ? Well if ya goal is to get on web site OC leader boards, then that's probably the best choice. OCCT and P95 will certainly put a heavy load on ya PC, but no application will ever place that load on ya PC again so I don't see the point. As an analogy, in New Zealand, a person with a certain skin type may want to limit themselves to say 15 minutes of direct sun exposure since they sit right below the hole in the Ozone layer. That doesn't mean that everyone in the world needs to limit there exposure to 15 minutes. The other problem is that with many sytnhetics, it's one dimensional and doesn't simulate a multitasking environment.

Having used almost all of them, the one thing I can say is that synthetic stable OCs fail under a multitasking benchmarks like RoG Real Bench Suite which is 4 tests (Image Editing / H.264 / Open CL / All of the preceding at same time) that run about 2 minutes each, the last one is the one that usually brings the box down. If I can get thru those 8 minutes, I know I am very close. The other thing is the RB Suite puts a bigger load on the system and this puts more strain on the PSU / MoBo power delivery system and perhaps the power swings have some impact on noise / voltage stability and therefore system stability.

When trying a new setting, I run the 4th tests ... if I get thru the 2 minutes, I run all 4. After 10 minutes adjustments are down to the 2nd decimal point. A final run of 4 hours is enough to lock it in.

My box serves as a combination SOHO Office and Multimedia Server, CAD Workstation, FlightSim and Gaming Box. In short.... what does your PC need to do to warrant a more "stressful test". A Soccer Mom's Van doesn't need to go pass the Baja 1000 to get the kids to Little League practice. To my eyes, anything over OCNs Medium category is fine for any workstation / desktop / gaming box. From OCN's Overclocking Guide. OCN recommends H264 (MediumCategory) as the Go To OC Stress test. Real Bench has h264 as one of the four modules but it's the last multitasking one that usually trips up most OCs.

"Quick Word About "24/7 Stability and Safety"
24/7 stability is a request people often make without thinking about what it actually means. '24/7' refers to the amount of time a CPU is spent under load. It says nothing about the type of load. Playing video games a couple of hours a day or week is not the same as hammering your CPU at 100% load for hundreds of hours in a row. If you're really that concerned about CPU longevity you shouldn't be using Prime95 to stress"

"Marathon-Man:
OCCT
Linpack (Max) (From Intel's website, not from OCCT or any other place or XTU.)
P95 28.10

Tough:
P95 27.9
IBT (Max)

Medium:

x264 16T
ROG Real Bench

Easy:
Stockfish (Chess, BMI2 version)
XTU
Aida64 (Full Suite)

Walk in the Park:
Cinebench
Firestrike
Booting into Windows


The other thing is ... what temps do you want to expose ya system to. Here's a comparison of the temp impact of the various stress tests for CPU temps and Power
http://www.overclock.net/content/type/61/id/2598037/
http://www.overclock.net/content/type/61/id/2574943/

Some numbers from above links
Prime95 28.7 @ 8k (Marathon) =68.5C / 187 watts
OCCT (Marathon) - 61.5C / 178 Watts
AIDA Full Suite (Easy) = 47.9C / 142 watts
RoG Real Bench (Medium) = 56.1C / 330 watts

Now if ya limiting ya OC tpo a certain temp.

But the best way to learn more is to pick whatever gets to your goal w/ the shorted amount of effort. Ocassionally I will just pick one of the other utilities out of the "Stress test Utilites folder and run it for "Shitz and Gigles" to see if anything can trip it up. This box is just over 4 years old and it doesn't crash under everyday use .... BF3 gave it sissues for a while till they got the coding issues sorted. Rocket League has ocassional issues but they come and go with each update. And much better now than when 1st came out.
 
Back
Top