How to guide: Memtest86+, Prime95, and SP2004

[John]

I figured that I would compile a little guide that explains how to properly run Memtest86+, Prime95, and SP2004 (A GUI version of Prime95) on a single/dual/quad core system to test for stability. If any of these programs generate an error, or other stop related messages, it generally means that you're unstable.

My dual core overclocking routine

I start out with Memtest86+ test #5 for 5-10 passes. If it passes I'll run SP2004 for 5-10 min to see if it errors. If it passes I'll run SP2004 for 12-24 hours. If it passes then you should be stable.

Important: If you get errors check the usual settings in your BIOS (cpu & memory voltages, FSB/HTT, timings, etc.) and try again. Keep your full load temps below 75C if you're using CoreTemp since the readings are typically 10-15C higher than bios/speedfan/everest reported temps. If you need additional how-to overclock advice please spend some time sifting through the threads in this forum.


Download the files

• Memtest86+
  Prime95 25.6 32-bit
  Prime95 25.6 64-bit
  Stress Prime 2004 - single core
  Stress Prime 2004 - dual core
  Intel Thermal Analysis Tool (TAT) - Not for AMD/P4. Core support only.

Memtest86+
[*]the floppy or bootable iso will work fine
[*]boot system to floppy or cd and Memtest86+ will start
[*]I have found that test #5 is the one that it will usually error on. On your keyboard hit C, 1, 3, 5 (enter), 0
[*]If you can complete 5 passes of test #5 try to complete at least two full passes of all default tests without errors
[*]Hit Esc when you are finished, remove the media, and boot to the OS

Please keep in mind that memtest is not the final verdict on stability. You should run it if you are testing an overclock, adding new memory or cpu to an existing system, troubleshooting OS errors that lead to memory failure, or if you have a new build.

Stress Prime 2004 (SP2004)

The GUI interface to Prime95's Torture Test which is much easier to use. Blend is the default and recommended test.


.: Single core system :.

SP2004
[*]run the .exe
[*]choose Blend
[*]click Start

Prime95 (use only if you don't wish to run SP2004)
[*]run the .exe
[*]choose Just Stress Testing
[*]choose Blend

.: Dual core system :.

Open your Task Manager (crtl+alt+del) and you should see two blocks under 'cpu usage history' with the cpu usage at 100% when running these utilities.

SP2004 Orthos Edition
[*]run the .exe
[*]choose Blend
[*]click Start

Prime95 (use only if you don't wish to run SP2004 Orthos)
[*]run the .exe
[*]choose Just Stress Testing
[*]choose Blend

.: Quad core system :.

Prime95
[*]run the .exe
[*]choose Just Stress Testing
[*]choose Blend

Orthos x2
[*]open Windows' Task Manager
[*]change the CPU Affinities of one of the Orthos processes
[*]when you switch back to the modified Orthos process, it will rescan its CPU Affinities and adjust the CPUs it will use accordingly.


Thermal grease application

Here's a pic :camera: showing the recommended method for newer cpu's. You'll find similar info on Arctic Silver's website.


Comments

There are other stability testing & burn-in apps available. Feel free to discuss your methodologies below.

 

[Kwint Sommer]

Combine this with a basic Overclocking guide and make it a sticky.

 

[AkumaX]

two stickies and call it a day

 

[Just4Ever]

stickilicious

 

[orangat]

Burnk7 is better than prime95 in generating heat and detecting undercooled cpu's.

http://pages.sbcglobal.net/redelm/

 

[videopho]

John
I see your Vcore at 1.5v. Is that a little high for x2 dual core? What's your temp @ load at 2.7ghz? Look at this linky. Any idea?

 

[John]

1.55v is probably the threshold for the X2; at least that is all I would be willing to push it. My full load prime95 torture test temps are 52C w/ a room temp of 78-80F (hot in West Texas). I use an XP-90 w/ a low rpm Vantec stealth fan and several 80mm Panaflo L1a case fans. My system is pretty quiet. Anything below 65C is ideal IMHO, and since I am under 55C I have no worries at all. :beer:

 

[Powermoloch]

It's nice to see a guide for a change

I did almost the same thing with your setup. 12 hours on memtest86+, 12 hours on prime95 and 32 M on super pi. At least I know my sempron 3100+ loads under 47 C

But I'm puzzled on one particular game...RTCW:ET. If it froze...I have to give a hard reboot, and the motherboard beeps out saying there was no Memory detected. Although it happens once in a blue moon, I'm still worried. Yet, I can play my other games without a problem such as oblivion on several few hour sessions.

Sempron 3100+ @ 2.069Ghz w/ Mushkin Blue XP4000 2 x 1gb sticks @ 230 mhz 1:1 ratio

 

[videopho]

Memtest86 question...
Does Memtest86 test out L1/L2 cache memories too? Someone please enlighten me as to why bothers to run memtest86 in regards to cpu upgrade?

 

[John]

http://www.memtest86.com/#trouble

Please be aware that not all errors reported by Memtest86 are due to bad memory. The test implicitly tests the CPU, L1 and L2 caches as well as the motherboard. It is impossible for the test to determine what causes the failure to occur. However, most failures will be due to a problem with memory module. When it is not, the only option is to replace parts until the failure is corrected.

 

[videopho]

John
Thanks for the response. It looks like my x2 runs into a brickwall @2.4ghz in which it failed or froze during Super-Pi's 32m calculation. 3dmark06 seems to run okay at that speed so did Prime95 (for >12hrs torture test as recommended). Memtest86 also passed all 5 tests as recommended. That being said, so at 2.4ghz the only that failed was Super-Pi I in which I may conclude that the system is still considreably unstable.
It seems to run fime @ 2.32ghz though. The only thing I could think of is either the mobo or psu or both which might have been the culprit to such a weak overclock with my x2.
EDITED: I did raise my vcore to > 1.4v and Dram volt to 2.79 v. Temps are 39-42c @ load

 

[John]

videopho, your temps look great. You could try 2.4GHz @ 1.5 vcore, 4x HTT, and drop your memory bus to 166MHz @ 2.6v cas 3-3-3-6

 

[aldamon]

Originally posted by: videopho
Memtest86 question...
Does Memtest86 test out L1/L2 cache memories too? Someone please enlighten me as to why bothers to run memtest86 in regards to cpu upgrade?

Memtest Test #5 is a great test for CPU stability. Like the OP, I use Test #5 as the first step in stability testing. It saves a lot of time.

 

[Mucker]

Nice guide John

Did you see this: http://forums.anandtech.com/messageview.aspx?catid=28&threadid=1904898
It seems to be much better than Prime95. I like it....

Ha...I'm at 1.60v on my X23800 (2620 MHz) on an Epox 9NPA+ Sli. It's Orthos SP2004 stable. Temps do not exceed 57C on both cores using a TT Big Typhoon. I'll report back if and when she fails although I suspect I'll be plenty fine until my Conroe upgrade in about a year....I have good reason to believe the Epox sensors are fuxored....

m

 

[Smartazz]

At what point to processors begin to get damaged? Around what temperature?

 

[Comdrpopnfresh]

Originally posted by: Mucker


Ha...I'm at 1.60v on my X23800 (2620 MHz) on an Epox 9NPA+ Sli. It's Orthos SP2004 stable. Temps do not exceed 57C on both cores using a TT Big Typhoon.

HA! My X2 3800+ runs a 2620 as well, but at 1.36volts! I believe that is 15% lower than your. BTW. I read that you shouldn't ever exceed 1.5 volts on an X2....
I idle at 44*, and max... well not sure there. lol

At what point to processors begin to get damaged? Around what temperature?

Depends on the processor... check AMD's website
http://www.amd.com/us-en/assets/content...e/white_papers_and_tech_docs/30430.pdf

I've seen better links, a complete comliation of each model. Sorry i couldn't find it. Damage can happen at any temperature if the voltage is too high. But I know for my X2 3800 I'm not supposed to go above a core temperature of 70*C, andi'm not supposed to go above a voltage of 1.55V. these are obviously not the ceiling of what the chip can go to, but the floor for where you might experience failure, or instability....

 

[John]

Originally posted by: Comdrpopnfresh
HA! My X2 3800+ runs a 2620 as well, but at 1.36volts! I believe that is 15% lower than your. BTW. I read that you shouldn't ever exceed 1.5 volts on an X2....
I idle at 44*, and max... well not sure there. lol

44C idle is kind of steep for 1.36v, but there are a lot of fluctuations in socket thermistor temps. For reference I am at 32C idle, and the temp rises 20C under full load. Therefore you're probably hitting 64-65C which is way to hot for your low core voltage. You might want to consider cleaning the old thermal grease or pad off and evaluate your current cooling.

 

[NoStateofMind]

I downloaded Prime95 and tried to edit the target line, but it keeps telling me it's invalid. I would really like to use it for my C2D. Any help?

Thanks

EDIT: Got it figured out

 

[Vegitto]

Good guide, John!

John for elite!
Oh wait..

 

[KeithTalent]

Thanks very much for this guide, it is very helpful!

I ran all of the tests as described and I got a BSOD when running Prime and SuperPi (the second time I ran it). All of the Memtest tests were passed, including the 5 passes (actually did 7) of #5.

So does this mean my RAM is dying? It is only a couple months old and my system has never been overclocked in any way.

Are there any other tests I can do?

Thanks in advance for any help!

KT

Edit: actually I get "NOT CONVERGENT IN SQR05" when running SuperPi.

Edit2: I just failed Prime95 (only ran single instance of it) and failed, no BSOD though.

 

[John]

Try raising the memory latencies a bit and bump the voltage to 2.7v (if possible). Be sure to stress test again.

Thank you all for the kind words.

 

[KeithTalent]

Originally posted by: John
Try raising the memory latencies a bit and bump the voltage to 2.7v (if possible). Be sure to stress test again.

Thank you all for the kind words.

Hmm, ok, please bear with me for a moment as I know next to nothing about this kind of stuff.

I have this RAM: 2048 MB of Corsair XMS PC2-5400 DDR2

Current timings on it are: 667Mhz 5-5-5-15-2T (what would be a safe bump for that?)

The only voltage settings I saw only went from 1.80 - 1.95 in .05 increments.

Sorry to be a noob, but I would really appreciate your help!

Thanks, KT

 

[Rubycon]

Something else I've noticed with super pi is non linear times. For example My FX60 at 3.0GHz takes 28 seconds for 1M places, 11 seconds for 512K places and 4 seconds for 256 places. Why is that?

 

[n7]

Originally posted by: MS Dawn
Something else I've noticed with super pi is non linear times. For example My FX60 at 3.0GHz takes 28 seconds for 1M places, 11 seconds for 512K places and 4 seconds for 256 places. Why is that?

I am not sure why, but AFAIK, SuperPI has never given linear results.
I suspect someone with more mathematical expertise could explain this.

BTW, this is indeed an excellent guide.

I personally have never liked Prime95 TBH, as i have found it to generally be a joke for testing stability.
I know with tweaking it can be setup better, & i know some have good success with it, but i see no point in fighting with problems when SP2004 takes care of all that.

SuperPI is fun for testing performance & very basic stability, but i wouldn't count on it for serious stability testing.
Dual 32M is okayish for finding issues, but still, not a great test overall.

Memtest is okay for testing RAM, but that's it.
It's rather useless for OCing i find, as running it doesn't test stability in Windows, & i think most people run their systems with their OS running

I used to use S&M as an alternative, but with the issue of their program reporting failed L1 cache on C2Ds, i haven't bothered with it lately.

To get to my point, i find that overall, SP20004's Orthos is by far the best overall stability indicator (excluding graphics of course) out there right now.
Results are extremely consistent, which helps tremendously when you like to do fine-detail tweaking like i do.
It's a hard program, meaning it will not tolerate sky-high unstable OCs like SuperPI, but for someone who wants a true stability test, i find it can't be beat.

Another couple for gamers i'd like to suggest are 3DMark05 & 06.
Both are very strenuous on the GPU, but also the CPU, & of course, the RAM.

If you can pass them, you'll be able to play games with most likely no stability issues (usually even when Orthos fails).

 

[tran1981]

How come Orthos blend gave me errors after 7hrs? I ran Orthos small FFT for 17hrs and memtest for 2hrs (6passes) but they didn't give me errors at same settings.