prime95 and memtest really the best stress testers??? Come see what I saw.....

[Duvie]

Ok I have seen this a couple of times now on the Albatron Pe mobo as well as my epox 4g4a mobo....


I have a speed like 2.4b@3.2ghz...prime tested for 5+hours (default settings) w/ 1 hour ran with 3dmark2k1se looping...memtest 10+ passes with all test and cache on....no bootup or restart issues.

I then go to run real world benches and I see that in Gknot (Divx encoding) will crap out (usually within 1-10minutes with an access violation involving avisynth.dll or virtualdub.exe....even ran with older programs and older versions same thing...

What solved it in every case???? Raise vcore by .025v...problem solved ran (2) 2hour encodings no errors....


What gives???


I also notice that divx was actually taxing the vcore and +12v rail much harder then the prime95 was....

Happened again at 3.24ghz and again a raise of .025v over what was prime95/memtest stable as above fixed any error....

saw it happen on the 4g4a at lower 3.15ghz speed...check out old thread and you will see me mention it....


All I have to say is maybe we are putting to much faith in prime95....run it with real world apps like a good encoder program that really crunches...

 

[LastRide]

I was always suspicious of prime.

 

[Megatomic]

Originally posted by: Duvie
All I have to say is maybe we are putting to much faith in prime95....run it with real world apps like a good encoder program that really crunches...

I don't think everyone out there has video files to encode or maybe even the technical knowledge needed to perform testing of that caliber. For the average gamer (not DiVX jock) Prime95/memtest86 may be the best bet.

Of course that's not to say that the DiVX encoding isn't more rigorous, maybe it's just not as practical.

 

[Duvie]

Autocadd would show signs of it on my 4g4a....Also Besweet MP3 encoder showed signs of it...I don't think the mp3 encoder is too intense...

Another thing I want to point out....With my 1.6@2.75ghz I would fail prime95 at 1.71v in about 15min yet I could do a 3hour divx encoding....

Why is it so sensitive now. I was using older program of prime95 but was told with C1 chips I should get the newer one. Divx is usually pretty error forgiving as it shouldn't crash but basically do a crappy job with video anamolies....


Because of this my testing regiment is....for final stability check...

1)5hours of prime95 (1hour w/ 3dmark looping)
2)10+ passes of all test in memtest (w/ cache on)
3)Divx encoding

If it passes those I feel very confident...

 

[Duvie]

I want to add this anamoly as we may call it right now only started showing itself at 3.15ghz and above on the 4g4a mobo and Albatron pe mobo....other then that what was prime stable was good enough for the benchmarks. However I was still running default vcore up to 3.15ghz so maybe at 3.06ghz I could have ran .025v lower and thus couldn't see if this pattern still followed suit...

 

[Jgtdragon]

I agreed. I ran memtest that tested all 11 test and passed couple time at 167fsb, but my chip won't boot into Windows. That tells me that memtest test the memory and cache of the cpu, but not the motherboard/chipset.

 

[SpideyCU]

Yeah, I've had memtest86 miss big problems with memory before. It made me suspect the CPU as the culprit, but when replacing the processor didn't help, I tried the memory even though memtest said all was good (and ran overnight just fine). Switching the memory was the trick - sometimes a little more digging is required. I basically view these kinds of programs as "OK, if something goes wrong here, then it's DEFINITELY this piece of hardware's fault", but if it passes, then it's only a "maybe" - they don't get off that easily!

 

[gooseman]

Personally, I think that if your machine will run the programs YOU want, your games, video editing or whatever you do with your computer, and not crash on you then it is stable enough. That's just my opinion though, and we all know the saying about opinions!

 

[bgeh]

Duvie: How do you run with all 11 tests with cache enabled?

TIA

 

[bonkers325]

Originally posted by: LastRide
I was always suspicious of prime.

its not 100% load
found out hard way

 

[Duvie]

Originally posted by: bonkers325

Originally posted by: LastRide
I was always suspicious of prime.

its not 100% load
found out hard way

according to when I bring up task manager it is....It is just low priority and will take back seat to anything else that starts running..Also I don't think it taxes the sysem and chipset memory enough...

 

[Duvie]

Originally posted by: bgeh
Duvie: How do you run with all 11 tests with cache enabled?

TIA

Ok when you boot into memtest outside of windows...hit c for configurations...then the number 2 for test selection...then the number 3 for all test...then the number 1 for cache...then the number 2 for always on...then the number 0 for cancel to close menu....Now you are good to go....

I think standard test only does up to test 6 or 8, right!!!

 

[bgeh]

thanks Duvie

 

[First]

That's why you should run at least 24 hours and not just 5 hours of Prime95. I've encountered at least 2 845PE motherboards that failed P95 at around the 13-16 hour mark during FSB overclocking.

 

[Duvie]

good suggestion!!!! normally though with i845d and g mobos I would test for 5 hours and nver had an issue with it not being the clear determining factor of stability...

The questionable side of that is why 13-16hours of prime before failure can be duplicated in 10minutes or less with a Divx encoding program!!! I guess more sad then questionable.

One thing I noticed as mentioned above was that prime95 was not intensely heating up the sys and cpu cores like the divx encoding...In the past prime95 was my best source for seeing high temp under load. I think this NB chip on the albatron gts quite hot or at least reported so. I see it hit 39c under load, and 31 at idle...ofcourse this is a guess it is the sys temp but it is the lower of the two for the cpu...I wonder if I bring those temps down if it will be moe stable at lower voltage....

That sys temp of 31c is 30c with case side off and still goes up to 39c so case cooling is no issues and room temp is only 74-75f....This is obviously a heat transfer or reporting issue...



What do you think is the reason for this then and the pe chipset???

 

[RalfHutter]

I've seen Prime95 failures in the 15-20 hour range. I always run it for 24-72hrs, depending on how patient I am.

It does run at low priority by default, but you can easily bump it up higher in Task manager if you want. When I run Prime95 I keep Task Manager open to make sure Prime95 is always getting 95%+ of the CPU time. Sometimes something burps (maybe Windows Prefetch starts running or something) and it's usage goes way down. If this happens I just reboot and start all over again.

 

[MetroRider]

soemthing i've wanted to add for a while, but havent done so until now...

i have seen that super rock solid systems of mine may crap out in Prime95 anywhere from 20 minutes to 3 hours in testing, yet in every other program i use, passes with flying colors. the machine can run MemTest86 for 30-40 hours with no problems, and encode, decode, 3d render, and play every game out there, yet craps out relatively quickly in Prime95.

does this mean that my machine is not so 'stable' after all? or is it just a case of Prime95 not being the right tool in this case?

thanks again...

 

[Shimmishim]

that's a good question...

i like to run prime for at least 6 hours to consider it stable...

i found out the hard way...

one time i didn't prime it but the computer was stable... or so i thought... in the middle of a game, my computer rebooted...

 

[OatMan]

I've also found its kind of an art finding the "instability" with variouse tools.

Memtest found a problem that went away, when I swapped the memory out. I later found out that it was not the memory at all, but that there was an incompatibility of some kind with just that CPU and the first mem module. Kinda weird, it seemed that individually all of the components were fine, but it certain combinations, they didn't get "plays well with others" on their report card...

I guess testing is just a matter of giving yourself the best chance to find a problem as quickly and easily as poss. There is not a gauranteed solution. I've found Burn6 to be usefull in combination with others like P95, memtest, and Photoshop transforms, 3dmark etc. I think everyone finds their own voodoo.

 

[nealh]

I wonder about prime as well...I can not run Prime 95 on My 2.4@3.0 but ca nuse dvd2scvd to convert a dvd to mpeg file with no issue and video quality looks great....DVD2SCVD uses TMPEG/BSweet and really pushes the cpu...I can perfrom 3-4 hrs of encoding with no issues

 

[Duvie]

All good stuff.....

I think days of just running 1 hour of prime95 doesn't mean as much.....I also revamped my tetsing and did 12hours of prime at 2.4@3.24ghz w/ 450ddr with no errors....I am going to do 2.4b@3.33ghz w/ 370 as well....

 

[Spicedaddy]

I don't use memtest anymore because I don't think it's accurate. I know my ram craps out above 153MHz because it won't even load windows (tested at 4:3 so it's not the CPU), but it would test fine in memtest up to like 158MHz. I may be wrong, but I think it's meant more to test memory for hardware defects rather than stress testing it.

Prime95 has always been 100% representative for me, so I trust it. I usually let it run overnight with 3DMark at the same time. If all goes well, it's stable for me.

 

[Duvie]

Memtest does appear to test the stressfulness at the given speed for me....At 450ddr I can make it thru all the tests with 10 passes w/o errors....

At 452ddr I get errors in 2nd pass in test #8 and then get errors pretty much in the same test for the rest of the way....That test, tests block moves....

I can boot into windows all the way up to 460ddr but start to get program errors and ofcourse tons more memtest errors...prime95 still doesn't error at 452ddr and doesn't until 456ddr....

I don't know if I trust any of the programs solely...A collabortaion of all of them may be the most sound choice. Now that doesn't mean if all of them passes except one you can disregard it....

 

[Goi]

I've had the opposite experience. I've had gknot/VirtualDub doing hours of encoding with no problem, as well as numerous other benchmark stressers, like SpecViewPerf 7.0, 3DMark2001SE, 3DMark 03, Quake3 and RtCW, but Prime95 torture test will report an error within half a minute of running. 2 other programs that crash out with GPF sometimes are Unreal II and UT2k3, but that might be just because of buggy programming, because I've seen this problem with numerous other people with stock systems.

 

[Duvie]

Goi,

I have had that happen as well as I can gknot my old 1.6@2.76ghz with 1.71v and do a 3 hour encoding yet fail prime in like 15 tests....

The point may be what cpu and system are you running???

There may be credible fact to the c1 stepping chips and the advanced hardware prefetch keeping the program from loading the cpu 100%...I watch task manager for quite awhile and it said 100% the whole time but who knows.