Prime95 vs. Linpack Xtreme

[Regeneration]

The video above demonstrates the superiority of Linpack Xtreme versus the latest version of Prime95. The overclocked PC passed nearly 2 hours of Prime95's small FFTs torture test and yet completely crashed within less than a minute and a half with the bootable version of Linpack Xtreme integrated on Porteus Linux.

Computer specifications: Intel Xeon W3680 @ 4.17 GHz (144x29), ASUS P6T (vanilla), 3x 4GB G.Skill Ares F3 @ 2304 MHz (10-12-12-31 CR1), and Nvidia GeForce GTX 970.

 

[Batou1986]

I have been trying to tell people for years prime is not enough to stress a modern CPU for years but they wont listen.

 

[Regeneration]

It took 77 seconds for Linpack to hard lock the entire system

 

[steen]

You've got interesting ratios for a W3680. Is that just a max stable for P95? Mine are OEM so 4GHz 133x30 & 3x4GB 1333 ECC is P95/Linpack stable. Bclk >138MHz results in instahang. They've been running the last 4 years @ 3.8GHz all core turbo. With decent motherboards I'd be running bclk 166x25 & 1866 DDR3.

 

[Regeneration]

You've got interesting ratios for a W3680. Is that just a max stable for P95? Mine are OEM so 4GHz 133x30 & 3x4GB 1333 ECC is P95/Linpack stable. Bclk >138MHz results in instahang. They've been running the last 4 years @ 3.8GHz all core turbo. With decent motherboards I'd be running bclk 166x25 & 1866 DDR3.

Max stable for Linux.

 

[Space Lynx]

no offense but prime95 load is unrealistic for 90% of us gamers, so its stable enough for us.

 

[Regeneration]

no offense but prime95 load is unrealistic for 90% of us gamers, so its stable enough for us.

There is no such thing as 'unrealistic load'.

Computers were originally created to solve equations. And that's what Prime95 and Linpack do, just in a larger scale and for longer periods.

Even basic features like 'Windows Update' need to solve equations, check checksums, etc. Not to mention plenty of 3rd party apps, and even games' DRM.

 

[hat]

Could you run the Windows version of Linpack Xtreme and see how it stacks up against the bootable Linux version?

 

[Regeneration]

Could you run the Windows version of Linpack Xtreme and see how it stacks up against the bootable Linux version?

Linux crashes a lot faster (300%) than Windows due to the balanced load distribution across both physical and virtual cores.

 

[hat]

But shouldn't it be 100% load? 100% should be 100%, and load distribution shouldn't really come into play in such a situation... unless I'm sorely mistaken. I'll take your word for it, just curious of the mechanics. Does it also find unstable memory faster under Linux?

 

[Regeneration]

Linux treats all cores as physical.
Windows prioritizes physical over virtual cores.

And yes, unstable memory crashes faster on Linux as well.

 

[Mussels]

we need a torture test linux build with CPU GPU and ram stress tests built in

 

[kastriot]

If oc-ed system in windows passes prime for 12-24 hours and intelburn 128 passes @Extreme it means it's perfectly stable for 99.99% people who mostly play games or whatever and for people who use PC for bussines or science they do NOT oc their PC, so this test is practically useless because 99.9% use windows here.

 

[Vayra86]

no offense but prime95 load is unrealistic for 90% of us gamers, so its stable enough for us.

Speak for yourself, that whole argument is complete BS and you know it. This topic is about the fact that even Prime isn't really a nasty load to put on a CPU these days. Go figure. A CPU can't even do a mild stress test (but it can run 5+ Ghz in idle or with 10% load! yay!) and 'gamurs' call it a proper OC...

Skipping tests because your CPU is cooking due to it is a clear sign you should be backing off. Instability is a clear sign you should be backing off. Anything else and you're deluding yourself.

Linpack was always a pretty good test to run for stability, glad to see it hasn't changed

 

[Space Lynx]

Speak for yourself, that whole argument is complete BS and you know it. This topic is about the fact that even Prime isn't really a nasty load to put on a CPU these days. Go figure. A CPU can't even do a mild stress test (but it can run 5+ Ghz in idle or with 10% load! yay!) and 'gamurs' call it a proper OC...

Skipping tests because your CPU is cooking due to it is a clear sign you should be backing off. Instability is a clear sign you should be backing off. Anything else and you're deluding yourself.

Linpack was always a pretty good test to run for stability, glad to see it hasn't changed

hmmm well speak for yourselves, my old PC failed Prime95 Small FFT all the time, but I could still run 10 hour plus gaming marathon sessions with 0 issues and 0 downclocking. /shrug

 

[Vya Domus]

Each to their own, if you wish to stress your PC for days on end with absurdly intensive tests by all means go ahead. Just so you know, no matter how absurd the test is, it still can't possibly cover all instruction combinations and whatnot therefore a window for instability will always be there. If you think that you are definitely safe after you do all that then you are just as disillusioned.

Testing for stuff that is outside your use case it's simply useless, sure it may make you sleep better at night knowing that but it's still useless nonetheless. Even Intel or AMD had cases where they shipped CPUs that were unstable under specific situations and most CPUs today likely still have weaknesses like that buried deep inside obscure instruction combinations.

I'll stick with something like AIDA64 FPU test, perfectly reasonable. I am not launching satellites into space and calculating trajectories and neither are most of you. But like I said, whatever floats your boat.

 

[Mussels]

some of us dont want to risk crashes or issues ever, others are willing to.

Those who dont fully test their systems are the same ones who come crawling to the forums asking for help, because this one specific game/program crashes and not the others, or they corrupted a hard drive and need data saved...

 

[Space Lynx]

some of us dont want to risk crashes or issues ever, others are willing to.

Those who dont fully test their systems are the same ones who come crawling to the forums asking for help, because this one specific game/program crashes and not the others, or they corrupted a hard drive and need data saved...

very subjective and generalised statement, but sure.

 

[Regeneration]

hmmm well speak for yourselves, my old PC failed Prime95 Small FFT all the time, but I could still run 10 hour plus gaming marathon sessions with 0 issues and 0 downclocking. /shrug

There was a version of Prime95 that crashed on AMD Phenom II CPUs for absolutely no reason. Took me days to figure it out.

 

[er557]

Ok, just ran it, the default test passed fine, also then ran a 35000 problem size test, about most of it ran fine, while heating up to 89c, which is uncomfortable for me,
the results were 653 gflops,
nonetheless, when I also browsed the internet with firefox during the test, the pc rebooted, so I chalk it up to kernel instability or something gpu related, as standalone the console window ran fine.
Also, it didn't properly detect 36 cores, 72 threads, as didn't the sensors console.
Temps for both cpu's in the tray detected fine.
Specs: Non overclocked dual xeon e5 2686 v3, with hacked turbo @72 threads @3.5ghz, undervolted core , cache, and system agent(-80,-50,-50), with microcode 0x39 loaded on boot.

 

[EarthDog]

Each to their own, if you wish to stress your PC for days on end with absurdly intensive tests by all means go ahead. Just so you know, no matter how absurd the test is, it still can't possibly cover all instruction combinations and whatnot therefore a window for instability will always be there. If you think that you are definitely safe after you do all that then you are just as disillusioned.

Testing for stuff that is outside your use case it's simply useless, sure it may make you sleep better at night knowing that but it's still useless nonetheless. Even Intel or AMD had cases where they shipped CPUs that were unstable under specific situations and most CPUs today likely still have weaknesses like that buried deep inside obscure instruction combinations.

I'll stick with something like AIDA64 FPU test, perfectly reasonable. I am not launching satellites into space and calculating trajectories and neither are most of you. But like I said, whatever floats your boat.

+1.

P95, using a combo of small fft and blend (using versions with avx/fma/etc), is fine for plenty. More is overkill. Aida64 stress test + fpu only stresses my system fine. When I say fine, I say it passes for at least 6 hours (ea test). For how I use my PC, mostly gaming, photoshop editing, some encoding, works a charm. THAT, gents, is stable and can be found with existing applications.

Something more stressful than p95 is like running furmark on a gpu.... a power virus and overkill. And for gpus, just plain dumb.

Thanks but... no thanks. No need for users to go to this level.

 

[er557]

I think the Linux kernel is less protected or isolated than windows , which is why stress tests don't crash there. It's indeed an extreme use case scenario, it's like I don't test my car on the Monaco grand prix, rather on the city streets.

 

[Regeneration]

I think the Linux kernel is less protected or isolated than windows , which is why stress tests don't crash there. It's indeed an extreme use case scenario, it's like I don't test my car on the Monaco grand prix, rather on the city streets.

Professional mechanics use computer and a special garage environment to test cars.

Some hardware components use Linux to update firmware. You surely don't want your SSD firmware update to fail in the middle because of OC instability.

Hardware failures on Linux usually crash the entire system (kernel panic). That reboot you mentioned, probably had nothing to do with the GPU.

If the CPU runs fine on Linux with stock clocks, it should run fine even when overclocked.

 

[Vayra86]

some of us dont want to risk crashes or issues ever, others are willing to.

Those who dont fully test their systems are the same ones who come crawling to the forums asking for help, because this one specific game/program crashes and not the others, or they corrupted a hard drive and need data saved...

Amen to this.

When my rig crashes, I simply know for a fact the overclock is not the issue. I can then analyze the problem for what it is without having to reproduce it on stock settings.

That said everyone should do things however they like, I just think its wrong to defend a half-assed overclocking practice as a great way to do it. Its really not - the benefit of that extra 100 mhz doesn't ever really pay off anyway in meaningful performance, but the instability is a pain in the behind and can even cost you money when it happens at unfortunate times. And (my) experience tells me: its always an unfortunate moment.

 

[SIGSEGV]

For me, personally, prefer to use 3DMark and gaming that use a high load of resources to test the system stability. Do the test for 2 - 3 loop (spy, firestrike ultra or combination) and gaming for the rest.