Stress testing a CPU install?

[ss88]

Upgraded to 12900K. Is there a good test to run to confirm good application of paste and contact w/ heatsink?

Also, I'm using Thermalright Peerless Assassin 120 and kinda want to see how much I can push it before CPU throttles (not overclocking).

 

[Grebuloner]

Prime95

 

[Dreamerbydesign]

OCCT cpu stress test.

 

[wandplus]

3D Mark

 

[pendragon1]

all decent suggestions ^^ although with the prime95, look for 26.6

 

[ss88]

I ran Prime95 (couldn't find 26.6, so just downloaded the latest). Ran "Just stress testing" option for 15min, no errors or warnings in Prime95.

CPU is i9-12900K with ThermalRight Peerless Assassin 120 air cooler on MSI Z690 Pro A board.

Avg temp was 77*C. Max was 100*C on several, but not all, of the cores. There were some brief periods of thermal throttling.

All good, and sounds like no defects in CPU, board, cooler or in my application of thermal paste?

I did record HWinfo and Prime95 screens if there's anything in particular I should look for.

 

[hmz]

I am really surprised people still run these tests. I do think they are kinda pointless.

 

[freeagentt]

PA120 is a 260w cooler, probably not strong enough for a full load.

I like Linpack Xtreme, the 10GB load.

 

[Grebuloner]

I am really surprised people still run these tests. I do think they are kinda pointless.

I expect my CPUs/GPUs to run at 100% load for days at a time (because they often do). Stress testing is a fast way to check my tweaks/OCs and ensure that the cooler is doing a good enough job, or is at the right combination of temperature and loudness.

But I am not like most users.

 

[kalston]

Yea I don't really even overclock nowadays but I still use stress tests all the time on stock parts to fine tune my cooling solution, so that in regular use I will never have to worry about noise or performance, even in summer.

 

[hmz]

I expect my CPUs/GPUs to run at 100% load for days at a time (because they often do). Stress testing is a fast way to check my tweaks/OCs and ensure that the cooler is doing a good enough job, or is at the right combination of temperature and loudness.

But I am not like most users.

I've used the prime95 running for hours with no errors, but then the game would crash within minutes resulting in a bad overclock. Since then I've dropped the ball using the prime and just go straight to the game if no crash playing for hours- then the OC is rock solid.

 

[Dan_D]

I am really surprised people still run these tests. I do think they are kinda pointless.

They really are.

I've worked on and I've maintained 10's of thousands of servers, most with multiple processors. When I was a service tech in shops, I worked on hundreds if not thousands of computers. I've built computers professionally, for myself, friends etc. God knows how many of those I've done. Out of all those machines, I've seen maybe a dozen bad CPU's. Only two or three were DOA. A few degraded from serious overclocking and a few others actually failed at some point. Usually they were taken out by the motherboard. This was also far more common back when people had analog modems which would get fried in thunderstorms and take out additional hardware inside the PC.

Stress testing your CPU to determine if its stable or reliable is absolutely pointless. Any decently built system with a reasonably good quality motherboard, RAM and PSU should be stable even at default settings. Use your new machine and if its not stable, then break out the stress tests, logs, etc. to find out why its not behaving. Doing some 24 hour or 48 hour burn-in test is ridiculous. It's an old way of thinking and really wasn't necessary even back in the day unless you were overclocking. All you do is delay using your new PC for no reason. Its been my experience that if you are going to have issues with system stability on a new build, its going to be an immediate problem.

Doing reviews, if I am going to encounter a problem during stress tests its usually within 1 or 2 hours at most that shit goes sideways. 24 hours or more of these types of tests is overkill. I'd also like to point out that when doing board reviews for HardOCP, we saw plenty of systems that could crush OCCT, Intel Burn In Tests, Prime95, 3D Mark etc. but they'd fail inside of 5 minutes doing a Handbrake encode. I'm not saying that these tests have zero value, but the idea that you need to do a 24 or 48 hour burn in test on new hardware for stability is crazy to me and pretty much always has been. Doing one of these tests does not guarantee anything good or bad.

 

[OFaceSIG]

Agree with Dan_D. I've been in the server game since 2007. I ran both production datacenters for the biggest dating website in the world. I can't recall ever getting a DOA cpu myself. I've had maybe 3 fail in my entire life and I've been building x86 machines and working on them since 1989. The reliability curve for x86 machines and their associated RAM is a bell curve. They usually fail early on or when they are VERY old. Unless you're doing something extreme like LN2 cooled overclocking, they usually just run.

Don't bother stress testing. If something fails, it's probably going to fail in the first year or use or at like year 18. That's just usually how it works out.

 

[GotNoRice]

I am really surprised people still run these tests. I do think they are kinda pointless.

I can think of several reasons to "stability test", just off the top of my head.

-To verify if the heatsink/waterblock is properly mounted. If your CPU throttles instantly, then it might not be making good contact...
-If you under-volt to get better boost speeds, there is a threshold where you can undervolt too much and cause stability issues. Obviously stability testing helps.
-Similarly, If you use something like PBO with Curve Optimizer on an AMD chip and really want to dial it in, stability testing can help.
-You might just want to know how your cooling solution performs at full load to see what the acoustics and temps are like, to dial-in your fan curve or possibly swap fans.

I agree that stability testing is not necessary in the context of testing for "CPU defects", but even if you run 100% stock and don't tweak anything, you would still want to know if your heatsink/waterblock is mounted correctly. You still might want to know what your cooling solution sounds like and how it performs at full load.

 

[Dan_D]

I can think of several reasons to "stability test", just off the top of my head.

-To verify if the heatsink/waterblock is properly mounted. If your CPU throttles instantly, then it might not be making good contact...

This doesn't take long. You will be able to determine this within minutes to an hour.

-If you under-volt to get better boost speeds, there is a threshold where you can undervolt too much and cause stability issues. Obviously stability testing helps.

I'm not talking about stability testing while overclocking or cooling. Even then, you don't need 24 hours or more to figure this out. If its going to fail, it will do it within 3 or 4 hours. Usually it happens in far less time.

-Similarly, If you use something like PBO with Curve Optimizer on an AMD chip and really want to dial it in, stability testing can help.

Again, I'm not saying stability testing has no use. I'm saying that its not necessary with brand new hardware at default settings. Even if you want to check, you don't need a full 24 hour burn in test as is often recommended or done by by people on this forum.

-You might just want to know how your cooling solution performs at full load to see what the acoustics and temps are like, to dial-in your fan curve or possibly swap fans.

Again, this doesn't take long to figure out. These CPU's will hit their max temperature in seconds. You can determine whether or not your cooling is up to the task in probably an hour or so. It doesn't take a full 24 hours to do this.

I agree that stability testing is not necessary in the context of testing for "CPU defects", but even if you run 100% stock and don't tweak anything, you would still want to know if your heatsink/waterblock is mounted correctly. You still might want to know what your cooling solution sounds like and how it performs at full load.

Again, this takes a couple hours tops. It's not something that requires a 24 or 48 hour burn in test.

 

[GotNoRice]

This doesn't take long. You will be able to determine this within minutes to an hour.

Even then, you don't need 24 hours or more to figure this out. If its going to fail, it will do it within 3 or 4 hours. Usually it happens in far less time.

Even if you want to check, you don't need a full 24 hour burn in test as is often recommended or done by by people on this forum.

You can determine whether or not your cooling is up to the task in probably an hour or so. It doesn't take a full 24 hours to do this.

Again, this takes a couple hours tops. It's not something that requires a 24 or 48 hour burn in test.

I never said anything about 24 hours... OP said he ran Prime95 for 15 minutes...

Ran "Just stress testing" option for 15min, no errors or warnings in Prime95.

 

[pendragon1]

some testing should be done on a new build, especially if youre building it for someone, but typically i only do like 15-20min and then start using as normal. basically for the reasons gotnorice listed, but ill add; make sure the drives/ram are performing as they should. .