Is Prime95 harmful?

B

balnazzar

Weaksauce
Joined
Mar 13, 2013
Messages
121
Hi folks.
Got a Xeon 8260M upon a X11SPA-TF here, just finished to build it. It works, at least for now.

Now, I always used to test my processors with Prime95, latest version, half an hour of blend + FFT small.
But recently, a colleague of mine told me that Prime95 can kill stuff (either the CPU, the mobo, or the PSU), since it causes unnormal voltage spikes.
I'd like to hear your opinion about that. Did anyone of you ever killed some piece of hardware by using Prime?
I really need to test system stability since I'll use that rig to work. If it's not stable, I'll lose money and reputation.

Other components: EVGA 1000W Titanium, four RDIMM modules, three RTX 2060 Super.

Thanks.
 
Prime is an unusually stressful workload since it can leverage AVX 512, but I have never heard about unnormal voltage spikes being a concern. In my opinion you are going to be seeing the absolute worst case in terms of heat and power draw so your system power and cooling need to be up to the task, but this would be my only concern. Plenty of us in the DC forum side that run PrimeGrid (which is very similar) for weeks or months on end with 100% load and we do just fine. An overclock can be tested pretty quick to be problematic under these situations so most of us run stock or if anything we undervolt to keep power and temps in check. If anyone told me that "cpu "X" shouldn't be used for prime" I would consider it not to be a good product.
 
use 26.6 to avoid the avx thing, new versions might be able to turn it off, not sure. run it for an hour or so. its fine for limited use. then game and use it normally.
 
I use Prime as a burn-in tool to make sure my system is stable, OC or no OC. If a system can't handle Prime/Furmark/Linpack then I would be wary about doing anything besides browsing the web with it.
 
pendragon1 said:
use 26.6 to avoid the avx thing, new versions might be able to turn it off, not sure. run it for an hour or so. its fine for limited use. then game and use it normally.
Click to expand...

Why would you avoid AVX unless you have no intention of using it in your normal work loads. It is a feature of the CPU that greatly speeds up certain work... Sure I get it if you are a gamer wanting to test your overclock for stability and you don't run the kinds of stuff day to day that use AVX, but simply saying to avoid it makes no sense to me.
 
motqalden said:
Why would you avoid AVX unless you have no intention of using it in your normal work loads. It is a feature of the CPU that greatly speeds up certain work... Sure I get it if you are a gamer wanting to test your overclock for stability and you don't run the kinds of stuff day to day that use AVX, but simply saying to avoid it makes no sense to me.
Click to expand...
most dont and the avx versions do overwork the cpu, especially if an offset isnt set.
 
motqalden said:
Prime is an unusually stressful workload since it can leverage AVX 512, but I have never heard about unnormal voltage spikes being a concern. In my opinion you are going to be seeing the absolute worst case in terms of heat and power draw so your system power and cooling need to be up to the task, but this would be my only concern. Plenty of us in the DC forum side that run PrimeGrid (which is very similar) for weeks or months on end with 100% load and we do just fine. An overclock can be tested pretty quick to be problematic under these situations so most of us run stock or if anything we undervolt to keep power and temps in check. If anyone told me that "cpu "X" shouldn't be used for prime" I would consider it not to be a good product.
Click to expand...

I don't overclock. Besides, that kind of hardware cannot be overclocked (I think), but I appreciate your recommendation.
I just want to make sure I have a stable machine to work with, since it's new.
Now I did the blend test full steam with AVX-512 activated, then another half hour without AVX-512 (just AVX2). I got the Supermicro 4U Tower cooler (it's a 92mm fanned heatsink with rather generous fins, but nothing special). At no point the temperature exceeded 78C (but the damn contraption gets loud).

CoreTemp shows 187W maximum power draw during the AVX-512 stress test, and ~160W during the non-avx512 stress test. I think these readings are OK for a 165W TDP processor.
 
Last edited:
pendragon1 said:
use 26.6 to avoid the avx thing, new versions might be able to turn it off, not sure. run it for an hour or so. its fine for limited use. then game and use it normally.
Click to expand...
No, in truth I'd like to use an AVX stress test, since my workloads with data augmentation make use of AVX (albeit not 512).
 
Jonnycat99 said:
I use Prime as a burn-in tool to make sure my system is stable, OC or no OC. If a system can't handle Prime/Furmark/Linpack then I would be wary about doing anything besides browsing the web with it.
Click to expand...
Exactly what I thought. Thanks.
 
Last edited:
If the voltage is not set to fully auto and you let it get too high, any stress test can harm the CPU.
Its up to you to monitor it before committing to a long test to be sure.
Very high temps with over volting is damaging.
 
I've seen prime fail after 24 hours, and I've had crashes after it has passed 24 hours.. Prime 95 as a stress test is a waste of time.

Try Linpack Xtreme! :D
 
I don't use Prime95, but stressers do exactly that. They are designed to put your CPU under load. The fact is many people's systems are simply not designed to properly cool and/or power their systems under load. Just the way it is.

So, don't blame the tool for putting your hardware under stress. That's what they are supposed to do. If your systems does well under stress, you should be pretty confident about it's ability to perform long term without "anomalies".
 
UltraTaco said:
[Gasp] taco was living a lie all this time..😯
Click to expand...
I used to swear by prime, my cooling was for apps like p95 and the like.. insane load generators. It just let me down too many times. It became untrustworthy to me, nothing good to say about it other than it can get your CPU hot :D

I really like Linpack from TPU for things like that. Its quick, its dirty, and your cpu will beg you to stop if you have a good clock on it. I recommend some airflow for that.
 
freeagentt said:
I've seen prime fail after 24 hours, and I've had crashes after it has passed 24 hours.. Prime 95 as a stress test is a waste of time.

Try Linpack Xtreme! :D
Click to expand...

Which Prime test did you run? I've not had issues with Prime. Each of the different mods of Prime has a different role to play.
 
First of all, if running Prime95 someone breaks the system, the problem isn't with Prime95 but something wrong with the hardware/cooling. It's extremely simple software which does no more than make use of the CPU like any other software can. It will push the CPU with high loads but that's what it's supposed to do and the system is supposed to be able to handle it. It tends to be one of the best pieces of software to test your CPU cooling. If your CPU cooling is adequate to run Prime95 then it's likely adequate for any other CPU task you could possibly throw at it.

There is no one size fits all stability/benchmark test or software. Some software will only stress certain parts of the CPU. Other software will stress other parts of the CPU. Stress tests for CPUs are generally worthless for RAM testing so you need to find something for that. Neither CPU nor RAM stress tests tend to touch the GPU so you need something else for that. No single piece of software is going to properly stress every piece of hardware or even every single feature on a single piece of hardware.

Stress test software rarely finds issues with hardware because most hardware run stock is perfectly good although you will find the occasional DOA or broken item which is why you want to stress test. Most people find value in stress test software when attempting to run hardware past stock settings and this is where stress tests find the most problems because you're pushing hardware beyond what it was spec'd to run at.
 
thesmokingman said:
Which Prime test did you run? I've not had issues with Prime. Each of the different mods of Prime has a different role to play.
Click to expand...

I don't recall, but it was about 10 or 12 years ago. I think I might have tried a couple of years ago too. I didn't have a problem running it though. Not since I moved to Intel lol. All of the problems I had was when I was an AMD user.
 
Prime should be fine on a Xeon - as far as I know, Xeons are warrantied to run any conceivable workload for three years at max temps. Where it gets tricky is on consumer processors - a few years ago, Intel changed their official warranty policy such that consumer CPUs were only rated to last three years under 'typical client loads'. There was also a largely unrelated issue where some motherboard firmwares were putting excessive voltage into CPUs under AVX2, especially overclocked.
 
Always makes me laugh when I see folks say "your system isnt classed as stable unless you run Prime for 48 hours+ without a crash!" Well as long as your Dad is paying the electric bill! :D

It only has to be stable as long as you would normally run your system at 100% imo. If Prime stays fine for couple of 20 minute runs on my system, it's good. This approach has never done me wrong all these years.
 
motqalden said:
Why would you avoid AVX unless you have no intention of using it in your normal work loads. It is a feature of the CPU that greatly speeds up certain work... Sure I get it if you are a gamer wanting to test your overclock for stability and you don't run the kinds of stuff day to day that use AVX, but simply saying to avoid it makes no sense to me.
Click to expand...
As the prime and intel guys said - AVX at full power (all cores, all segments, if you don't have an offset programmed) is actually beyond the specs of the CPU in MANY ways, even at stock or near-stock speeds. The chips weren't designed to do that really - IIRC, they said it was effectively running the CPU @ 130% of TDP at stock, hence the offsets. Intel has even chimed in with that recommendation on overclocks, as the offset also does weird things to voltages.

There's almost nothing in the real world that uses AVX right now, especially with it being the entire workload (it's a small part of transcode/encode workloads) - it's mostly the space of oddball AI and ML workloads for the entire chip, I believe (feel free to correct me, been a bit since I looked this up). All HPC stuff trickling down.

https://forums.tomshardware.com/threads/prime95-no-avx.3515075/
Intel folk chime in on this thread.
 
That just sounds to me like intel pushes their chips too hard at stock and they can't properly utilize AVX at the boosts / frequencies they ship at. basically sounds like a lame excuse.
 
anyways test with AVX if you use AVX test without if you don't . use and offset if your chip is gonna catch fire.
 
You must log in or register to reply here.
Back
Top