Prime95 - How long is long enough?

TeeJayHoward

TeeJayHoward

Limpness Supreme
Joined
Feb 8, 2005
Messages
12,271
Just finished building a new SFF setup. 4080 FE, 7950X3D, AXP90-X47 Full Copper (with Noctua NF-A9 25mm fan sub), in a Fractal Design Terra. I'm running Prime95 right now to ensure that the cooler can handle the load. So far, it's hit 90*C and is doing it's thermal throttle thing, but not all the way down. I'm averaging 4-4.2GHz over the last hour.

I'm about to go to bed. Is an hour of Prime95 good enough to be considered "stable" or should I start logging and leave it running overnight?

Some pics for attention:
IMG_2609.JPG

IMG_2612.JPG
 
There are a lot of people on here that think you need 24 hours of Prime95 stability testing to call something stable. I'm going to go against the grain of that. I've done a lot of testing doing motherboard and CPU reviews over the years and I can tell you that more often than not, an hour is sufficient. If you are going to have a problem it usually happens in less time than that. Naturally, there are edge cases where someone encountered issues during a 24 hour test on hour 18, minute 37 or some such time. However, that's not often the case. I'm also going to take this a step further and say that Prime95 stability testing isn't the end all be all of tests. Back in the [H]ard|OCP days we found that some systems would pass Prime95 and other tests with ease and die in 2 minutes of Handbrake encoding.

I think an hour is fine to make sure your cooling is working properly and you don't have some memory setting causing you issues. Use your system, enjoy it and if you have any stability issues you can revisit the topic.
 
imho, it's not a good stability test. It's good for ensuring your overclock/settings aren't completely borked and checking your cooling, maybe.
 
Nobu said:
imho, it's not a good stability test. It's good for ensuring your overclock/settings aren't completely borked and checking your cooling, maybe.
Click to expand...
It will definitely heat up your CPU. That said, so do a lot of other tests.
 
^ Whatever the guys said said.
To add: What I do is run a few hours (1-2) of prime95 not just post-assembly, but, say, once a year, something like that. Or/and boot to memtest and leave it crunching overnight. Why not? It's good to know early on that something might be nearing its end.
 
The OCCT PSU test will get your CPU way hotter than P95 in less than 2min.
I usually run that for 1hr and I transcode a 4K 25000KB BluRay rip to 8000KB 1080p MKV using Handbrake.
If it can pass all of that, I'm usually golden 99.99% of the time.
 
Last edited:
Heat saturation will probably be reached within an hour. As for stability, I disagree with above posts and consider a 24 h run the minimum. Sure, *most* of the time an error will occur within an hour, or some other test will fail sooner, but enough times I've had overclocks fail 10-20 hours into testing. Those often had issues pop up during normal workloads even if days apart. It's either stable or it isn't and any difference compared to a stock state is unacceptable to me.
 
I do one pass and done. It just needs to run as long as you'd maybe thrash your machine for 100%. Which in most cases isnt that long.

I always used to laugh at posts here saying "Oh you need to run Prime95 for 1-3 days solid to class your machine as stable!"

Yeah you still live at home don't you! ;)
 
daglesj said:
I always used to laugh at posts here saying "Oh you need to run Prime95 for 1-3 days solid to class your machine as stable!"
Click to expand...
I just don't want to have my 15 hour render crash at the last minute if I can prevent it 🤷‍♂️
 
Meeho said:
I just don't want to have my 15 hour render crash at the last minute if I can prevent it 🤷‍♂️
Click to expand...

There is always that risk of something 'crashing' no matter how long you test it or run it. For most people they are wasting time and energy stress testing for more than 30 minutes to an hour.
 
daglesj said:
There is always that risk of something 'crashing' no matter how long you test it or run it. For most people they are wasting time and energy stress testing for more than 30 minutes to an hour.
Click to expand...
Yes, but with longer testing I've found I could go from 98% to 99,99% probability of being stable, which proved significant enough to be worth it. As I've mentioned before, I had Prime95 crashes 15-20 hours into testing that showed up as a BSOD once a week. Once I got it stable to pass 24, never had the BSOD again.
 
Lol I don't know about "prime stable" are any normal applications or games going to push the CPU equally as hard? Maybe some professional apps but certainly not gaming right?
 
xDiVolatilX said:
Lol I don't know about "prime stable" are any normal applications or games going to push the CPU equally as hard? Maybe some professional apps but certainly not gaming right?
Click to expand...
It's not a matter of how hard, but type (variability) of workload. As Dan_D said, a system could be Prime stable but fail during an encode or a "simple" game load. Proper stability testing includes both synthetics and real life use.

In my example, system was >10h Prime stable but BSODed during browsing or similar mundane tasks.
 
Or just run what you use the system for.
Ex. if you're a gamer, letting it run demanding games, timedemos, et al on loops for a day or two.
If you're doing NLE, create an automated task and let it run too.
Chances are if there are issues it will error out, reboot, freeze, etc. within that time.
Sometimes nearly immediately in the case of insufficient voltages.

A good blended test is the Asus realbench.
OCCT uses a blend of Linpack and Furmark as well.
All have their pros and cons.
I've had systems that failed STOCK too.
 
Used to run prime for at minimum a week. small FFTs will get things hot and loose...

But then someone put out a linpak front end and life changed... Nothing would make your chip glow hotter. Found instability where prime didn't... I think someone put out another frontend since then called linpak xtreme
 
cpufrost said:
Linpack
Click to expand...
Glad you mentioned it, been meaning to ask. What's the status on that one - how tough is it on things like Zen 2? I'm on Matisse and about to do a case clean-up, so torture would be nice.
 
FrgMstr said:
Go to sleep, start it. Wake up, check results.
Click to expand...
Instructions unclear.
Went to sleep, started it, woke up and nothing was running.
Went to sleep again, and my machine was now the Apollo 11 guidance computer with old NES controllers as the only input device and it was overheating.

I'll see myself out.
 
michalrz said:
Glad you mentioned it, been meaning to ask. What's the status on that one - how tough is it on things like Zen 2? I'm on Matisse and about to do a case clean-up, so torture would be nice.
Click to expand...
It's pretty hard for sure. Use HWINFO or similar to monitor individual core temps to reveal bad TIM, etc.

Prime95 with AVX-512 is absolutely brutal on Intel CPUs, I lost a VRM on an 18 core Xeon doing that! It's pretty unrealistic for 99% of the stuff out there, just like pushing a car with 2000 brake horsepower to the point where the rotors look no different than the ring of fire in the sky during an annular solar eclipse! ;-)
 
I'm with Dan_D and btw, Prime95 does not stress the 7950X3D (or any other X3D) at all! so for those CPUs it's probably not even worth doing more than a few minutes (it's because they have a lot of safeties built-in regarding voltage, amps etc to avoid killing the cache). You will heat up your CPU WAY more with other tests such as Cinebench R23. All the Aida64 short benchmarks are great for stability testing too, but the CPU stress there is also not very demanding on a X3D CPU.

You will literally get higher CPU temps in some games vs Prime95 with a X3D CPU, sometimes even with a single threaded load. Have a look here for an example.
 
Last edited:
I would usually go for 24 hours, but because I ended up so busy it would somethings accidentally run for 48+ hours before I was able to get back to it.
 
When I was doing sports overclocking I used 36 hours as the time to run mprime/prime95. Although I don't remember the details I must have gotten an error later than 24 hours.
 
For quick check, I usually use y-cruncher Pi-2.5b or 5b.
If you pass that, you're golden as it will test your cpu, imc, ram, and max heat your system can handle all at once.
 
Kinda pointless now. Most CPUs already pushed to the edge and not worth overlooking anymore. I just set PBO and move on with my life.
 
Prime95 is a waste of time. I have seen it crash after 24 hours. Not worth the electricity.
 
You must log in or register to reply here.
Back
Top