How can Prime95 report errors but my system be 100% stable?

[AluminumStudios]

I do a fair amount of reading on CPU arcitecture and understand a lot of the nitty gritty details.

But I am confused over Prime95's cpu tourture test. On my Athlon XP2200+ (tbred-a) @ 1930 MHz & 1.7 volts it errors out anytime between 10 minutes to 2 hours.

My system is however 100% stable. I let it encode mpegs and render from AfterEffects for hours and hours on end with no problems at all. It has uptimes of weeks sometimes between reboots. Reboots are usually because buggy software that I used decides to get bitchy or I install something new.

My question is how can my CPU have errors and not be functioning perfectly without causing Windows to blue screen or the system to hang, or at least do something like give me a miscolored pixel in something I'm rendering from AFterEffects or encoding to MPEG. It seems to me that the CPU (or memory) shouldn't be selective about what instructions it errors on and would sooner or later do something to bring my system down ... which hasn't happened.

 

[jaeger66]

If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

 

[Duvie]

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I agree...round it up with a memtest (all test through like 10 loops or so)....This is also stressful and can verify those errors....


Now I hear many with c1 P4's have had issues running older version of prime95 and giving false errors, but I haven't heard anything on the amd front...

 

[NFS4]

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

 

[AluminumStudios]

My point is if my cpu isn't stable then it must be doing something like flipping bits. It seems to me that any error the CPU has would be equally likely to affect instructions used in calculations such as Prime95 does or in normal system operations. And if an error occured during an instruction that was part of a normal system operation I should feel the effect of that in some way such as a blue screen, a freeze, a program crash. That never happens. My system is rock solid - even during some hellish AfterEffects renders that I put it though that involved all kinds of instructions because of a varity of algorithms used in the various filters and alpha-channel operations.

People are putting blind faith into Prime95 and their documentation takes a hard line approach that if it fails the hardware isn't perfect. What I'm trying to figure out is if there are perhaps factors that Prime95 doesn't take into account. Are there any precision differences between numeric instructions on the Athlon vs Pentium chips? I read in an interview with AMD before where they refered to giving one of their older chips "classic IEEE floating point." Is it that a CPU has to comply with a certain level of precision, but CAN be taken further than that? Could there be an instruction on the Athlon that Prime95 is expecting more precision?

To me an unstable or imperfect system is one that crashes, freezes, has unexplained errors. I take my sytem to through the ringer with AfterEffects, and when AE starts eating a GIG of RAM and crunching nested composits with dozens of layers each it's every bit as much of a workout if not more than Prime95 can give it.

I searched for Memtest on-line and started reading through it's documentation and you need to boot from a floppy disk to run it. Scr3w that, I don't even have a floppy drive.

So, can anyone tell me what really happens when Prime95 reports an error? Does anyone have a technical explanation beyond just saying the hardware isn't perfect?

 

[OS]

Originally posted by: NFS4

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

heh, no wonder I never get as high overclocks as other people.

 

[Duvie]

That is why I say follow it up with memtest...both test the cpu cache, memory sticks (thoroughly based on the amount you have, prime often you only allocate 8mb for the program), and memory subsystem....

You can always get it to boot off of a cd-rom!!!!


I can agree if you have stability then you are fine...however many programs and cadd based programs can be like this, can be very forgiving to errors which may be slight calculations resulting in sometimes unnoticeable to slightly noticeable errors in the rendered file...IE I could run divx on my old tbird 1.4@1.53ghz but I had a chance of having some slight pixel defects, to corrupted frame here or there, to a few bright colored dots around....prime said I had errors yet I never crashed....I reduced it to 1.5ghz and I never had it happen again on the system....


i can run divx encoding right now on my machine at 2.76ghz with 1.71v yet prime95, memtest, and 3dmark don't complete and report errors....


Check out prime95's site for FAQ...

 

[jaeger66]

Originally posted by: NFS4

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

It's just semantics really, the difference between "100% stable" and "stable enough for me". If your system does what you ask, then enjoy in good health. But if it fails Prime, it is not 100% stable as the original poster stated.

 

[NesuD]

AluminumStudios

I searched for Memtest on-line and started reading through it's documentation and you need to boot from a floppy disk to run it. Scr3w that, I don't even have a floppy drive.

You should have taken a trip over to the download page. There is a downloadable iso available for creating a bootable cd.
Download page

 

[merlocka]

Originally posted by: NFS4

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

Prime will flex corner conditions of timing which won't (probably ever) be challenged in most normal computing conditions.

Not passing Prime could still leave you with a system you believe to be "perfectly stable", however, "perfectly stable" only then means "It doesn't crash enough for me to care about".

 

[sharkeeper]

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

That's not a very good way to look at things. STOP overclocking EVERYTHING and run Prime95. Does it fail at the same point or can it run endlessly for days on end without producing rounding errors? If it still produces an error with NOTHING O/C then you have faulty hardware or a possible conflict that is causing this anomaly. If it runs error free with NONTHING O/C then the O/C is causing the instability. I hope you do nothing that is *critical* with your computer. Believe me, these kind of things will bite you in the balls like a 4 kg sewer rat ascending upon you on the throne when you least expect it. Definitely not worth the paltry "improvement" in performance IME/IMO.

Cheers!

 

[AluminumStudios]

I TOTALLY agree with NSF4. I'm not going to let Prime95 tell me that my system isn't perfect when it runs for weeks on end running intense Apps like AfterEffects which use up all of my RAM and send me deep into virtualmemory while keeping my cpu floored doing internet and intense fpu operations for hours on end.

Originally quoted by Merlocka:
Prime will flex corner conditions of timing which won't (probably ever) be challenged in most normal computing conditions.

Can you expand on this? I"m trying to understand in technical detail what Prime95 is detecting is wrong with the CPU. I can't imagine that an app should be dependent on any type of timing within a CPU seeing as how different CPUs have different architecures and it's the results of the instruction execution that is important.

I'm not going to beleive that my system is unstable until it show me by freezing, crashing, corrupting data, etc.

And I HAVE to overclock it because the operations I do are so slow that waiting long periods between renders before I can see the results then make changes and re-render my video destroys my workflow and productivity.

 

[DaveSimmons]

I TOTALLY agree with NSF4. I'm not going to let Prime95 tell me that my system isn't perfect when it runs for weeks on end running intense Apps like AfterEffects which use up all of my RAM and send me deep into virtualmemory while keeping my cpu floored doing internet and intense fpu operations for hours on end.

how do you know that your images aren't being slightly corrupted by 1 pixel here and there when a memory fetch or store goes slightly wrong? Working with corrupted data would not cause an effect filter to crash, it would just write an image slightly different from what it would write with the correct bitmap.

The same with games -- a texture read / write that's off by one bit wouldn't cause a crash, just a bad pixel that's gone by the next frame.

Remember the coppermine 1.13 fiasco? It ran "well enough" for intel to push it out the door, but tomshardware could make it consistently crash doing linux kernel compiles.

 

[NFS4]

Originally posted by: AluminumStudios
I TOTALLY agree with NSF4. I'm not going to let Prime95 tell me that my system isn't perfect when it runs for weeks on end running intense Apps like AfterEffects which use up all of my RAM and send me deep into virtualmemory while keeping my cpu floored doing internet and intense fpu operations for hours on end.

Originally quoted by Merlocka:
Prime will flex corner conditions of timing which won't (probably ever) be challenged in most normal computing conditions.

Can you expand on this? I"m trying to understand in technical detail what Prime95 is detecting is wrong with the CPU. I can't imagine that an app should be dependent on any type of timing within a CPU seeing as how different CPUs have different architecures and it's the results of the instruction execution that is important.

I'm not going to beleive that my system is unstable until it show me by freezing, crashing, corrupting data, etc.

And I HAVE to overclock it because the operations I do are so slow that waiting long periods between renders before I can see the results then make changes and re-render my video destroys my workflow and productivity.

Prime is damn screwey sometimes too. I was running my 2.4 @ the 156MHz bus @ 1.625V. Would get errors in Prime after 5 minutes. I dropped it down to 155MHz bus at 1.575V and got errors. Went to 1.6V and got errors. Went back up to 1.625V and got errors.

So I said screw it and left it at 155MHz bus @ 1.6V b/c all of my other apps were working fine. So this morning on a whim (after posting to this thread) I put Prime on at my current settings (2.4B @ 155MHz bus @ 1.6V) and left prime running in the background as I did my usual everyday stuff (surfing, Photoshop, recording TV, etc) and it has been going ever since without an error.

The temperature in this room is the same as it was the other day (maybe even a little bit higher) and both Wingznut and PM have said that "burning in" doesn't actually do anything so I chalk it up to Prime screwing with me.

Then again, my system also passed 10 passes (10 tests each) of MemTest yesterday and went through 8 hours of Sandra burn-in that same day.

Bottom line, I'm not going to let one benchmark tell me that my system is unstable.

 

[apoppin]

Originally posted by: NFS4

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

Wait a minute!!! Didn't you post in another thread that you had problems with BF1942 on your O/C'd system . . . or did I misunderstand you (hand hovers near "search" button)?

 

[NFS4]

Originally posted by: apoppin

Originally posted by: NFS4

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

Wait a minute!!! Didn't you post in another thread that you had problems with BF1942 on your O/C'd system . . . or did I misunderstand you (hand hovers near "search" button)?

On my old 1.8A @ 2.25 system with an Epox 4G4A motherboard. It wouldn't take my two sticks of memory without failing in BF1942. Prime would run fine, but BF1942 would lock up.

Another reason I don't trust prime.

My current system is a 2.4B @ 2.8GHz with a Gigabyte 8SQ800 using the same memory from the 4G4A system.

 

[apoppin]

Originally posted by: NFS4

Originally posted by: apoppin

Originally posted by: NFS4

Originally posted by: jaeger66
If Prime95 crashes, it isn't stable. It might be for what you're doing, but obviously you're not stressing it to the point of failure in everyday use.

I say f*&k Prime

My 2.4B B0 @ 2.8GHz stepping gets some errors in Prime ranging from 10 mins to 5 hours. However, my system is perfectly stable. Runs through hours of Sandra burn-in tests, passes MemTest, and can withstand me playing hours of BF1942 and Il-2. No to mention all of my DIVX encoding and TV recording.

Prime isn't the end-all be-all test IMHO.

Wait a minute!!! Didn't you post in another thread that you had problems with BF1942 on your O/C'd system . . . or did I misunderstand you (hand hovers near "search" button)?

On my old 1.8A @ 2.25 system with an Epox 4G4A motherboard. It wouldn't take my two sticks of memory without failing in BF1942. Prime would run fine, but BF1942 would lock up.

Another reason I don't trust prime.

My current system is a 2.4B @ 2.8GHz using the same memory from the 4G4A system.

Thank-you for clearing that up . . . see some of us not only read your posts, they remember them (somewhat)

 

[VTrider]

Interesting thread, and interesting notions of what people consider 'stable'.

This bothers me too, I run an overclocked Athlon (XP1600+ @ 1.753MHz) which can pass Prime95 for hours on end. If I bump it up just a bit and/or increase memory timings my system will still run days on end without crash, but will fail Prime95 in minutes. I don't mind a dropped frame in a game here or there, but I also do a lot of video editing, mpeg converting and I always wonder if that would ever screw up something important?

I have a hunch that these errors go unnoticed (sp?) from the end-users point of view, but the thought of messing of a SETI WU drives me nuts! Does hardware/software have redundant features built in, in case of errors like this?

-VTrider

 

[sharkeeper]

And I HAVE to overclock it because the operations I do are so slow that waiting long periods between renders before I can see the results then make changes and re-render my video destroys my workflow and productivity.

No you don't HAVE to O/C. O/C is a hobbyist/enthusiast activity. Anyone that is participating in O/C actions on live production equipment is taking unnecessary risk. If you need more processing power you distribute the jobs across multiple computers and or use SMP in workstations. Imagine if NASA started to O/C computers! :Q

I overclock computers on my dime and time. Just like I may modify my vehicle to for better acceleration, cornering, etc. However I would never do this to vehicles provided by the Administration. (regardless how funny it sounds!)

Bottom Line: If any single program becomes unstable after an O/C, your system is running in a condition that leads to failure or anomaly. The goal to a "successful overclock" is to achieve stability that is 100% identical to non O/C conditions but at a higher clock rate. Once you start breaking down you should back off.

Cheers!

 

[AluminumStudios]

I don't OC live or production equipment, I OC my own personal equipment for my own personal projects that I do for Anime conventions and such. I also don't have the budget to buy any more or larger hardware than I have.

I'm a systems analyst and run everything by the books at work.

Also, the goal of OC'ing is to get as much performance as possible from my hardware (with the fact that crashes, etc. are understood to be anti-performance.) I'm not going to trust one flaky app (Prime95) that tells me that my rock solid system that crunches video all day is unstable. Rather I believe Prime95 has the problem and people are putting too much blind trust into it. Especially since NO ONE has been able to say anythign other than "if it fails you have a hardware problem." No one (not even the PRime95 documentation) can give me a technical description of the nature of the failures that could trigger Prime95 to cause an error.

Also, if my system is erroring, I want to know why the effects have never been visible at all - no visible miscolored pixels on my screen, in my renders, no crashes, no blue screens, no random characters in my text files, no nothing that would indicate any type of error. If my memory or cache, or cpu is flipping bits somewhere it won't discriminate between video data and isntructions ... it will just happen and be equally likely to damage all data flowing through my system (including data and instructions - causing notable instability which DOESN'T EXIST!)

 

[Duvie]

Have you followed it up with memtest yet??? Go check out Thugsrook specs for memtest and go by that...It test cpu cache as well as physical memory and memory subsystem of the mobo...If you error there as well then it would be 2 sources. If not then dono't worry about it if you think it is stable and not affecting your work....

Like I said above some programs will crash and others will mask the errors into whatever you are doing...

 

[apoppin]

You never did mention if your system can run Prime95 without errors at default clockspeeds . . . well?

 

[AluminumStudios]

Duvie: I did run memtest and it passed just fine overclocked. I only did one pass of it becasue I had work to do and didn't have time to let it run. I'll let it run multiple passes overnight sometime soon.

apoppin: I need every last ounce of speed I can get ... I really don't care about default speed, just my overclocked state. If you used AfterEffects to you'd understand ... the pain of waiting 20 seconds per frame when redering a complex composite ....



I'm more concered with getting an answer to EXACTLY what happens when Prime95 fails and why it's an indicator of a problem when nothing else in the system fails.

 

[apoppin]

Originally posted by: AluminumStudios
Duvie: I did run memtest and it passed just fine overclocked. I only did one pass of it becasue I had work to do and didn't have time to let it run. I'll let it run multiple passes overnight sometime soon.

apoppin: I need every last ounce of speed I can get ... I really don't care about default speed, just my overclocked state. If you used AfterEffects to you'd understand ... the pain of waiting 20 seconds per frame when redering a complex composite ....



I'm more concered with getting an answer to EXACTLY what happens when Prime95 fails and why it's an indicator of a problem when nothing else in the system fails.

All I am suggesting is that you downclock your system to default and run Prime95 over night.

IF it fails . . . perhaps there is some problem with Prime95, it's installation or some other flaw in your system - that would tend to substantiate your claim.

On the OTHER hand - if your default system runs Prime95 flawlessly - there just might be something wong with YOUR theory and your O/C'd system.

Personally, I believe you have already made your mind up that your system is "stable" and would find it hard to accept "evidence" to the contrary.

 

[Duvie]

as far as I know the program is running existing already found prime numbers and is doing a mathemtical formula and then comparing it to already correctly figured number.....In the grand scheme of it all it is a simple process that the cpu may be failing to do....

What exactly is the error??? Is it a rounding error?? If so that is usually a cpu failure as it fails to get number exactly as it should have following the mathematical formula it is given in the algorithm....If it is hardware error I have seen that to be other items ranging from mobo chipset to ram....

What are you temps at full load??? I can render large files as well in my programs and even do 2-3hour divx encodings and prime95 has always been the program that can build my cpu temp the highest...maybe heat is a factor...


I though of another past experience...at 1.6@2.66 with 442mhz ddr w/ 1.69v I could pass prime for 12hours yet I would have restart issues and fail memtest with errors....the fact in this case is that I know that prime95 is not a definitive tester of physical memory....This is one example where even a passing prime95 test needed to be verified with other tests that do a different (not better) job at testing a certain area of a component...