CPU and Memory Pass Memtest Failed Prime 95 Who do I Trust

[agenda2005]

Hello,
My CPU/memory passed memtest (Test 1-8) in 15 pass and test 9 in 3 pass but failed prime 95 after 2hours 53min. Who do I trust? Can someone explain which of these tool is better for testing system stability.
The set up for the test is:
AMD Athlon 64 3500+ 130nm @ 228*11 = 2508MHz (IHS removed)
CPU Vcore = 1.6V
Abit AV8 rev 1.1
VDIMM = 2.8V
HTT Voltage = 1.3V
SB @ 2.6V
NB @ 1.6V
HTT *4 228*4 = 912MHz HT
Crucial Ballistix PC3200 @ 228 2.5-2-2-8
Thermalright XP-120 with 110cfm, 120mm Panaflo at full speed
Room temperature = 20C
Minimum CPU Idle temp = 35C, Maximum CPU load temp = 50C

[Ref]

You should trust BOTH. You know ... Prime95 counts 1+1=2 /well it's a bit more difficult, but it is not important right now/ and if your CPU counts 1+1=1.8613654 then something is WRONG with your CPU / RAM. It checks results against KNOWN results and if they are wrong, something failed and it reports the failurie.
HOWEVER, AFAIK Prime 95 23.7 had some error with 1 GB ram configurations, so if you are not using the latest version you should upgrade to latest to see if problem persists. If yes, then you surely have a hardware failurie.

[CarpeDiem]

I don't think it is a hardware failure, i think it has to do with his temps, if it was indeed a hardware failure, then it would have failed within the first hour, but he almost got 3 hours of prime stability, so maybe temps were getting high and caused an error, try playing games and running apps, if they don't crash then you don't really need to worry, if they do then downclock the fsb.

[agenda2005]

I think my temps are fine. I removed the IHS on the CPU so the Thermalright XP-120 with AS5 is right on the core and the room temp rattle around 17-20C(max) here. Night temps can go as low as 15C. The maximum temp that I saw was around 50C. Bios version 15 of Abit Av8 MB now calibrate the CPU temps correctly but I notice that the BIOS temp is always 5C greater that uGURU utility that monitors the hardware. At the most, I should have 40C idle and 55C load. I'm currently at work, once I get back home I will check the version of Prime 95 that I'm using to confirm what Ref was saying as my system uses 512MBx2 = 1GB memory in total.

[st0nedpenguin]

if it was indeed a hardware failure, then it would have failed within the first hour

Not at all, I've had a CPU fail prime95 after 24+ hours with the prommi evap head reading -50, so it's not necessarily a matter of temps, just because it doesn't error within the first hour of Prime95 doesn't mean it's stable. ;)

[IamAnoobieCheez]

I don't think it is a hardware failure, i think it has to do with his temps, if it was indeed a hardware failure, then it would have failed within the first hour, but he almost got 3 hours of prime stability, so maybe temps were getting high and caused an error
temps increase cause to failure is a hardware failure.

Hardware failed due to whatever reasons: temps & voltage fluctuations/spikes.

so yes, as heat builds up, it will simply worsen the situation. Voltage fluctuation is there, and heat goes up, the system will become a lot more sensitive. At this point some degree level of voltage fluctuation will be enough to lose stability.




As for checking true stability of the system, gaming(intensive ones) is the tool. There are systems out there can handle prime95(cpu & memory both) for 12~ 24 hours and yet cause random failure in games. This is where voltage fluctuation "can" often be the cause. The way I look at it, gaming is more intensive than memtest and Prime combined. This is with the video card running at default clocks, no overclocks.




try playing games and running apps, if they don't crash then you don't really need to worry, if they do then downclock the fsb.
Yes, I agree....
I would also give it *plenty* of time playing games, not just for one day, but at least several months.... If the system passes without any hiccups in that time frame, the system "is" stable, or should I call it, the true stability~~~

[agenda2005]

I agree that temps or voltage fluctuation can cause hardware failure, but temps should not be an issue in my case. What temp is too high for 0.13um Athlon 64?
I would be using my sytem mainly for scientific computing and memtest and prime 95 are what suite my situation better. Again my version of prime95 is 23.8. I need my system to work with good precision especially with floating point numbers without returning bad round off errors. I do a little gaming and video encoding, but they are not my primary concern. Who cares if a gamming session crashes, just back down your system and fire it up again. What is more of concern to me is for my sytem not to give bad results after a day or more time of computation.

[Crankster]

The Crankster defenition of a stable PC: If it doesn't crash enough to annoy me i'm good.

[reject]

i try prime, it fails, i up voltage(s) and pray it wont crash again when folding
my cpu is stable at my latest oc. i hope. hasnt crashed for a while. oh god i hate the feeling in my stomach i know im gonna jinx it
i have prime 23.7.1 is that 23.7?

[existz]

it does'nt really come down to what temp is good for your cpu, take my chip for example, (mobile 2400) its been water cooled for 5 months or so @ around 37c load , and i redone my setup and had a large bubble and it shot my temps up to 50c and it crashed @ windows, but once that was fixed and back down to the high 30s low 40s it was fine, just dose'nt like "high" temps. ive seen some chips that operates fine @ 55c or so but other dont come out of there shell until they are down in nice cold temps

[jlccarv]

i try prime, it fails, i up voltage(s) and pray it wont crash again when folding
my cpu is stable at my latest oc. i hope. hasnt crashed for a while. oh god i hate the feeling in my stomach i know im gonna jinx it
i have prime 23.7.1 is that 23.7?

looks like it... update to 23.8, as it does not have a problem with 1GB of memory...

[ugp]

From what I hear....with A64 don't trust Prime95. If MemTest86 is completely error free then test your system...just running it 24/7. If you don't come across any problems then most likely you are fine.

[Ref]

From what I hear....with A64 don't trust Prime95. If MemTest86 is completely error free then test your system...just running it 24/7. If you don't come across any problems then most likely you are fine.

But remember, if prime 23.8.1 (the newest version) FAILS, then you HAVE a HARDWARE FAILURIE. You should NOT use such system for mission critical applications. The system is maybe stable, but processor IS making errors.

[Smurfy]

My theroy is that memtest only tests your memory, not your CPU, now we know the A64 use built in memory controllers, so why the CPU isnt strained the memory controller can handle the high fsb's but as soon as it is it will fail earlier.

I can get 300+ FSB @ 2.5, 3, 3,7, but i cant get teh same clock speeds as lower fsb's cause prime will fail in minutes

[agenda2005]

I'm currently primming now with my A64 3500+ with crucial Ballistix PC3200 at 215*11 = 2365MHz (2-2-2-10) and no error so far after 20+ hours. My sweet spot is 2500MHz but I felt 2400Mhz is more reliable 24/7 for a 2200MHz rated CPU. I have done 219*11 = 2409MHz (2.5-2-2-10) and it primed sucessfully in dual channel after 12hours+. I couldn't get my memory to run in dual channel with 2-2-2-10 after 215MHz. One of my memory will do 222MHz and the other 216MHz at 2-2-2-x timing. Too bad. I need a replacement to stay on a low latency timing in dual chanel at DDR438. Working on getting a cherry picked crucial ballistix now.
Thanks Ref, I will not use it unless I have about 48Hours of suceessful prime95 and at least 50 sucessful loop of memtest. I will also include super pi 32M and CPU Burn in for 24Hours. I'm working on double precision floating point numbers that involve several iterations and any error will be blown out of proportion.

[ugp]

Actually a few of the tests in MemTest are very CPU related and will throw errors because of the CPU and not the memory. There is a guide somewhere I found that tells what each test actually is for...

If I find it I will post a link...

[bottone]

Personally, I've had prime95 pass the Small FFT option for at least 18 hours at a certain cpu voltage/speed, and if I run the in-place Large FFT it will fail consistently within a half an hour (could be specific to my setup though - a64 3000+ & shuttle an50r).
I've also had memtest+ pass at least once completely through, but testing with goldmem ( http://www.goldmemory.cz ) will freeze or error alot. You might want to try alternative methods, but I'd definitely say that if any one errors you do infact have some hardware problem.

[reject]

when i memtest at 2.5ghz, i will get a critical stop and it wont finish the testing.
i can pass superpi 32m at 2.4, no errors, but prime fails in 1/2 hour to hour.
im gonna get the new one but i dont knwo if ill run it cause i really dont want to drop speed and my chip is almost at 50c load.