Everest Vs Prime95

thegoonden

thegoonden

thegoonden

Associate
Joined
21 May 2007
Posts
1,464
Just wondering if you lot would consider Everest's system stability test to be a reasonable simulation of some hard computer use?

The reason is that I can run it for hours at 3.5GHz with the vcore sitting at 1.28 under full load, with temps peaking about 60. But to get prime95 happy at the same speed, it takes a vcore under load of 1.34ish with temps around 65-67.

Obviously I'd like to have the lower power, cooler chip. And nothing else seems bothered on the lower setting only Prime's small FFT test.

Should I just leave it low until something acts broken, or should I aim for Prime stability at all costs?
 
Prime95 will expose insufficient CPU cooling and computer case cooling, or excessive Vcore and overclock. At no other time will a CPU be as heavily loaded, or display higher temperatures, even when OC'd during worst-case loads such as gaming or video editing. Prime95 can be used with SpeedFan to observe CPU temps, while stress testing for system stability.
 
Last edited:
Pretty sure the problem is actually undervolting. As I said, 3.5GHz on 1.28v is taking the p*** a bit, but everything bar P95 trots along nicely.
I think, especially since it's warm weather, I'll just leave it and play it by ear. If other stuff starts bailing, I'll up the vcore a bit.

(cooling is a properly seated TT120 with arctic silver 5, and a Yate Loon H instead of the default fan, in an antec900 with fans on high but routed through controller, cooling and airflow are not too far up my list of worries :D )

BTW, correct me if I'm wrong, but coretemp/realtemp read their values directly from the chip do they not?
If so either the Yate, the AS5 or my new Gigabyte mobo has scrubbed about 5-10 degrees off former temps (depending on situation)


The trouble is the Gigabyte has bad vdroop without LLC, and still allows a little with LLC on. So getting to prime nice, which needs 1.31/1.32 under 4 core full load, needs a bios setting of 1.4, which gives idle voltage of 1.36/1.37 ish, and idle temps in the mid 40's. Hardly excessive and it only pushes load temps up a teeny bit. But I've become a voltage and heat miser I think ;)
 
Well, things continue to be odd

The machine will run games and apps, as well as perpetual Everest testing, in perfect contentment at 3.5GHz, on v's of.........
1.36 BIOS setting
1.312 Idel Windows
1.296 Heavy load Windows
1.28 Full 4 core load Windows

But wont sit happy in Prime, until I up Vcore so the Full 4 core load reads at 1.32.

It's odd since on the previous board, this simply didn't happen, either the system was viciously unstable....if it couldn't prime, it'd fail in games, and often sitting on the desktop doing nowt, or rock solid for all things. It took a little more voltage too, though required a lower BIOS setting as the ASUS's LLC eliminates Vdroop almost entirely, whereas the Gigabyte simply reduces it.


Anyway, just in case anyone cares, I'm going to stick with the lower setting until it proves flakey as the higher voltages are pushing the idle temps into the 40's and load temps a touch too close to 70 for my liking, albeit this is at least partly due to the warm spell and an ambient temp of 28 in the afternoon. The chip seems a little less heat tolerant as well, compared to the ASUS.

I know I'm rambling about this a little, but the variations in behaviour that differing motherboards exhibit are quite interesting (I think anyway, and if I can't find someone HERE who agrees, I really am a very very tragic geek LOL )
 
Prime95's stress test will analyse your computer's ability to calculate known Mersenne prime numbers correctly. Where this doesn't happen, you get errors.

Your computer is effectively not calculating correctly. Whilst Prime tests - unsurprisingly - its ability to calculate or analyse prime numbers, such errors could be extrapolated onto things like vectors for 3D work/games, to burning a CD or even toying with encryption. CPUs do have a degree of error protection and compensation within their architecture to the everyday noticeable effects of such a problem may not be manifest. But when you think of the potential cumulative issue as a result of extended bad writes to cache, then memory, then hard disk, you have a potentially catastrophic problem (in terms of a small element of a file corrupting to something larger in scale).

Whilst the problems caused by incorrect CPU logic may not seem to pop up on a daily basis, they will be occuring: but who notices an incorrect pixel shade when playing a game, or the odd incorrect bit on a burned DVD or file element? You probably won't, but your computer may introduce so many as to wreck part of a file's structure or its data.

Everest? I'm not familiar with that application, so don't know how it works and therefore can't comment on it from a stress-test POV.

What you should bear in mind is that just because something fails Prime doesn't make the PC unusable, but just make sure you don't going BOINC-ing or Folding because your CPU cycles will be wasted. You may go through a lifetime of computing with you PC and not notice an error, or you could have a spectacular failure tomorrow.

The only way to make sure a computer is 100% stable is to test every single element of its functionality. And that takes a very long time and is wildly impractical for pretty much everybody. Prime is a good indicator of a CPU's ability to calculate prime numbers or factors: it is NOT a good test for 3D games, for example.

However, I suspect that in these days of easy overclocking, people want a quick and dirty way to achieve results and wave willies and Prime/Orthos is the accepted benchmark. But it is only really one side of a complex polygon.
 
But at least with prime on small ffts you can pretty much isolate a failure in the system to be that of the CPU..
Once you have that stable.. in prime, then you can move on to other tests and other software to stress other parts of the CPU separately.
I'd go as far to say if it isn't stable in prime at the current clock and you cant / don't want more heat produced then you only have one option, lower the clock 5Mhz or so.

I have found there is something going on with the software of Everest that makes my MB drop some of it's sensors completely, lord knows what it's actually doing.
Also noticed Everest does not stress the CPU as hard as prime, even with all boxes ticked.

Get it stable in Prime at whatever clock that requires IMO, then move on.
 
Yes, but why do I care if P95 doesn't run?

If it's never ever crashed in any game or app, surely logic dictates......

"it crashes of I run this".
"don't do that then".

End of.


As for Everest not testing as hard, that was my original question, is it a more reasonable simulation of heavy load?

Cubase has not yet pushed the cpu into double % usage figures. No game runs tiny calculations on all four cores.
The more I think about it, I'm just burning my chip out for a program that serves no purpose other than to burn chips out.


If it DOES start to go funny, I know precisely what to do.
 
Oh BTW, it's ONLY the small FFT that does it. It ran 22 hours on Blend.
Considering the machine is normally used only for an hour or two (I have a proper OS for desktop use other than games and cubase), I think that's good enough.
 
Everyone is different and it really depends on what you consider stable. But for me, if my overclock fails prime then it is not stable, even tho it might run games and apps fine.

It may run blend for hours, but that really stresses more of the ram than the CPU. If it was me i would be making sure it's stable in prime small FFT, just to be sure.

If you don't want to use more Vcore, then just lower the clock.
 
Last edited:
I suppose if I were interested in searching for large prime numbers it may be a problem.

First time Cubase/GPL/COD4/Crysis reboots, I'll up the voltage.

Kinda weird to my mind, setting standards based on a program that you don't use.
 
A failure in Prime95 means your CPU has made a calculation error. This indicates it *might* make an error under other circumstances. But if your normal apps perform without error, just run it as it is.

Although that does beg the question; why would you perform stability tests if you ignore the results?
 
On my previous system, as stated, Prime95 was a good indicator of stability, if it failed, everything else failed. Since changing motherboards, everything except prime works at lower voltages and temps. So on this particular system it is an extremely poor indicator of overall stability.
3.5GHz on 1.28v is great, and I WILL let you know if anything other than prime's small FFT ever crashes. Not happened yet.
 
thegoonden said:
On my previous system, as stated, Prime95 was a good indicator of stability, if it failed, everything else failed. Since changing motherboards, everything except prime works at lower voltages and temps. So on this particular system it is an extremely poor indicator of overall stability.
3.5GHz on 1.28v is great, and I WILL let you know if anything other than prime's small FFT ever crashes. Not happened yet.
Click to expand...

As we've all stated, it is of no consequence if you consider your computer to be functioning appropriately. However, your computer is not calculating correctly. That is all Prime95 says.

The reason why so much is hung off Prime95 is the myth that - if a PC is stable according to Prime (i.e. complete cycles without error) - then it will be stable full stop. This is not the case and has never been. Prime95 has only ever tested for CPU calculation errors together with those stored in cache/memory.

As systems have become increasingly power-hungry, it is increasingly important to test overall system stability, i.e through compilation of programs, encoding media files, playing games, rendering, etc. No one single test is sufficient to adequately demonstrate PC stability.

You wouldn't buy a second-hand car based on the fact that its engine started first time, would you? There could be a whole host of other problems with it. By the same token, checking the entire car is free of fault is impractical from both a skills perspective and one of time.

Personally, I would not ignore Prime95 errors because of my experience in neglecting such output and the consequences that ensue as a result, but that's because I've had critical applications die on me at key points. What difference is makes to you is down to you and you alone. Is the extra voltage going to make any difference aside from personal pride and CPU stability? You're not really going to get any more noise, are you?
 
The reasoning is this........
More Volts=more heat
More Volts+LLC is more strain on PWM apparently.
Therefor
More Volts=cpu/board dying younger.





Like I keep saying it's totally different on this board from the last where a prime failure meant other things would fail too.

What Prime has showed me is that my system is 100% stable under full 100% 4 core load for an indefinite time, PROVIDED those cores are calculating something other than tiny fast
fourier transforms.


The wisdom here seems to be ignore everything that tests your CPU, even if it loads it to 100% flat for days on end, if it cannot do very small FFT calculations, your machine is junk.

Now if someone can assure me that more volts and heat have no effect whatsoever on motherboard and CPU life, I'll go for it, but for now, as I've stated I will lay off the volts until any single process shows any single error.
 
thegoonden said:
The reasoning is this........
The wisdom here seems to be ignore everything that tests your CPU, even if it loads it to 100% flat for days on end, if it cannot do very small FFT calculations, your machine is junk.

Now if someone can assure me that more volts and heat have no effect whatsoever on motherboard and CPU life, I'll go for it, but for now, as I've stated I will lay off the volts until any single process shows any single error.
Click to expand...

*Sigh*

Nobody's is saying your CPU is junk/rubbish/pointless/futile/etc. All we are saying is that your CPU's apparent inability to calculate Prime factors might cause further problems for you. If that's a risk you want to take, that's fine with everybody here. If it makes you happy, then that's great, too.

With regards to CPU/motherboard life, there are no guarantees either component won't break tomorrow, or the day after, regardless of overclocking or otherwise. It is merely logic that dictates increased voltages and subsequently heat going through a CPU and a motherboard will have some impact on the integrity of that system over a certain period of time, but nobody has ever (to my knowledge) quantified this effect.

Now, assuming the mean time to failure of a chip is five years (almost positive it is significantly greater than this, purely because the only chip I've ever had that died was because the waterblock fell off - and this was when PCs didn't have heat-related shutdown... every other CPU I've ever experienced, has lasted longer than five years flawlessly), overclocking might reduce that. Nobody knows by how much, only that it might.

And to be honest, running 1.34 through that motherboard and its associated cooling will probably not even make your system bat an eyelid. New motherboards have 6-10 phase power regulation: back when I cranked my chips up to ~150W output, the motherboards had - at best - 3-phase regulators. And I never had a problem.

But all hardware is different. My current system may die tomorrow and it's underclocked. Yours may survive for another ten years... and you might even replace yours in two years' time.
 
Well, as it happens, I did some more testing, now that the weather is a bit more normal.
With the increased vcore (1.344 idle, 1.312 p95 FFT), and a decent ambient temp, the CPU barely hit 60.

And also, my own logic beats me........
Fair enough, P95 small FFT is the only thing that needs that much voltage, but then it's also the only thing to make the CPU hot enough to worry me.

Still, the different requirements of the two motherboards are kinda weird and interesting.
 
You must log in or register to reply here.
Back
Top Bottom