Orthos Stability Testing - Priority Settings + Length of Test

[txtmstrjoe]

I'm presently testing an overclock on my Opteron 165 (CCBBE 0616XPMW), running at 2.727GHz (303*9).

Before overclocking, I determined maximum levels for my board's FSB, CPU, and RAM. I also tested the RAM with MemTest86+ (v.1.70), and then used Gogar's AMD OC Optimizer. I've run S&M (at Long Test Length, Full Priority Testing) CPU + RAM test, and passed that. SuperPi on both cores simultaneously also passed. I'm now running Orthos, and have had to raise VCore from 1.3625V to 1.40V (raised systematically and incrementally), after the tests failed at around 3 - 4 1/2 Hrs.

I'm running Orthos at Priority 10, and I'm going for a minimum of 12 Hrs.

Is there a correlation between Priority Setting, the length of the test, and actual functional stability?

Thanks for your thoughts, people!

 

[jjb0070]

It seems to be a matter of taste. I think almost everybody would be satisfied with a full priority test in orthos run successfully for 24 hours. Any amount of time less than that, and it seems everybody's got an opinion, and they're all different!

 

[txtmstrjoe]

Thanks for your thoughts, jjb0070.

As it is, my rig hasn't yet made it past the 5-hour mark on a full-priority Orthos run. The temps are maxing out at circa 52-54 degs C for both cores at 1.4V VCore, so I'm not comfortable with giving it more juice.

I suppose 2.655GHz (295*9) is the stable best OC I'll get from this Opteron. Not too shabby, but I was hoping for at least 2.7GHz just on air.

Thanks again!

 

[nealh]

funny if you read Prime 95 they say to run at priority of 1..persoanlly if you are doing nothing else when stress testing(which I do..walk and let it run)....I see no reason to change the priority....

Prime95 and all stress testing is a funny thing..I found in the past if I was stable for 12hrs plus on large fft..I would have no issues on my system

others recommend 12 -24 hrs of each small, large, blend testing...
I just ran a C2D 12hrs blend stable when I stopped....

many feel blend stress everything ..cpu, ram, mobo etc...I have notice blend put alot stress on ram, cpu and chipset..is one stressed more not sure but isolated small fft seems to push a higher cpu temp than blend and early runs of large fft ..which is odd to me..my opteron large fft pushed max temps???

I have seen many say there seem is orthos/prime 95 stable and crashes in games or some other app

unfortuantely no stress is perfect but I would stress for at least 4-8hrs on large fft....small fft is primarily cpu, large seems to at least stress some ram and cpu

 

[Noubourne]

The thing is, if you already used MemTest 86 to test your RAM for stability at this clockspeed, then it's a waste of time to use Blend. If you're too lazy to run MemTest86, then I suppose using Blend is a cheap way to get some minor testing done on your RAM stability. I wouldn't count on it though.

The idea with these stress programs is to put your CPU through more stress than you would see with normal computing so that you are stable when doing just about anything you could throw at it. If you're including RAM you're not getting the full stress on your CPU that you would with CPU-only testing. I think blend is folly for CPU stress testing, because it takes stress off the CPU.

Small FFT is the way to go for Optys.

For C2D you can use large FFT b/c it still fits in the CPU cache.

Before I do Prime95, I'll usually do 3dMark01, 03, and 05, and the 3 most stressful games I have at the time to test for all-around stability. Prime 95 is my last test because it takes the longest.

 

[nealh]

Originally posted by: Noubourne
The thing is, if you already used MemTest 86 to test your RAM for stability at this clockspeed, then it's a waste of time to use Blend. If you're too lazy to run MemTest86, then I suppose using Blend is a cheap way to get some minor testing done on your RAM stability. I wouldn't count on it though.

The idea with these stress programs is to put your CPU through more stress than you would see with normal computing so that you are stable when doing just about anything you could throw at it. If you're including RAM you're not getting the full stress on your CPU that you would with CPU-only testing. I think blend is folly for CPU stress testing, because it takes stress off the CPU.

Small FFT is the way to go for Optys.

For C2D you can use large FFT b/c it still fits in the CPU cache.

Before I do Prime95, I'll usually do 3dMark01, 03, and 05, and the 3 most stressful games I have at the time to test for all-around stability. Prime 95 is my last test because it takes the longest.

ok..I will retest large fft on my C2d..I did find it interesting that for me small fft gave higher temps immed. on my C2D than large fft

so is small fft not a good choice for C2D...because cache is not stressed enough....

getting pretty confusing now

 

[Noubourne]

Stability testing is largely a subjective endeavor.

You should run whatever overclock you are comfortable with. These methods are intended only to give you an idea of what problems you might encounter with a specific overclock. None of them are the end-all be-all of stability or reliability. You are overclocked. There is some risk. The idea is to lower the risk so that it is as close as possible to the risk you would encounter at non-overclocked speeds.

These tools just help you to judge how far you are pushing it. If they all run without errors at stock, then they should all run without errors at your overclocked speed for lowest risk. If you get another 500Mhz OC with errors in only one testing tool that doesn't seem to affect your mission-critical applications, then who's to say you're not stable enough?

The goal is to be confident enough in the stability of your system that you do not think it will go down under stress of some mission-critical computing. For me this would be gaming or video editing/encoding. I'm not telling you to go back and re-stress overclocks you consider to be stable. I'm just telling you what other people are doing to stress test their own overclocks.

If it was my C2D, I'd probably do 10+ hrs on small and 10+ on large - but only if they gave me significantly different temps. I'd want as much of the cache tested as possible, but I'd also want some testing time done on the highest temps that I am likely to see.

Just my .02

 

[nealh]

very nice post..I agree 100%...

I was going to run a longer test with small fft becasue of temps but felt your opinion on large fft maybe more valid...I will probably run both to see what happens

 

[InterHmai]

oho thank you for posting this. I wasn't sure how to interpret Orthos failures and whatnot, especially with "stability" seeming so subjective so many different people.

 

[txtmstrjoe]

Originally posted by: Noubourne
Stability testing is largely a subjective endeavor.

You should run whatever overclock you are comfortable with. These methods are intended only to give you an idea of what problems you might encounter with a specific overclock. None of them are the end-all be-all of stability or reliability. You are overclocked. There is some risk. The idea is to lower the risk so that it is as close as possible to the risk you would encounter at non-overclocked speeds.

These tools just help you to judge how far you are pushing it. If they all run without errors at stock, then they should all run without errors at your overclocked speed for lowest risk. If you get another 500Mhz OC with errors in only one testing tool that doesn't seem to affect your mission-critical applications, then who's to say you're not stable enough?

The goal is to be confident enough in the stability of your system that you do not think it will go down under stress of some mission-critical computing. For me this would be gaming or video editing/encoding. I'm not telling you to go back and re-stress overclocks you consider to be stable. I'm just telling you what other people are doing to stress test their own overclocks.

If it was my C2D, I'd probably do 10+ hrs on small and 10+ on large - but only if they gave me significantly different temps. I'd want as much of the cache tested as possible, but I'd also want some testing time done on the highest temps that I am likely to see.

Just my .02

Perhaps I can seek you advice.

On said full-priority Orthos Blend test, the machine would sometimes simply shut down. The symptom rears its ugly head at almost always the same point in time (right around 4 1/2 to 5 Hrs. from the start of the test). I monitor the temps with CoreTemp, and the consistent maximum reading is about 53-54 degs C, so I'm not sure if the cause of the shutdown is heat-related.

I've ruled out the PSU as a potential cause as well; I've run these tests on both a Corsair HX620W and a ThermalTake ToughPower 750W, and CPU-Z and BIOS-level voltage monitoring reveal nothing alarming.

The aforementioned RAM testing was also done with MemTest86+ meticulously, and I'm at the maximum tested stable RAM frequency with the appropriate amount of VDimm (2.75V; higher or lower throws up MemTest86+ errors). Maximum tested RAM frequency was 248MHz.

What may be causing the shutdowns?

(As an aside, running lower-priority -- i.e., 1-5 -- Orthos testing, the machine passes the 24-hour limit easily with no errors.)

Thanks so much for your thoughts!

 

[nealh]

Originally posted by: txtmstrjoe

Originally posted by: Noubourne
Stability testing is largely a subjective endeavor.

You should run whatever overclock you are comfortable with. These methods are intended only to give you an idea of what problems you might encounter with a specific overclock. None of them are the end-all be-all of stability or reliability. You are overclocked. There is some risk. The idea is to lower the risk so that it is as close as possible to the risk you would encounter at non-overclocked speeds.

These tools just help you to judge how far you are pushing it. If they all run without errors at stock, then they should all run without errors at your overclocked speed for lowest risk. If you get another 500Mhz OC with errors in only one testing tool that doesn't seem to affect your mission-critical applications, then who's to say you're not stable enough?

The goal is to be confident enough in the stability of your system that you do not think it will go down under stress of some mission-critical computing. For me this would be gaming or video editing/encoding. I'm not telling you to go back and re-stress overclocks you consider to be stable. I'm just telling you what other people are doing to stress test their own overclocks.

If it was my C2D, I'd probably do 10+ hrs on small and 10+ on large - but only if they gave me significantly different temps. I'd want as much of the cache tested as possible, but I'd also want some testing time done on the highest temps that I am likely to see.

Just my .02

Perhaps I can seek you advice.

On said full-priority Orthos Blend test, the machine would sometimes simply shut down. The symptom rears its ugly head at almost always the same point in time (right around 4 1/2 to 5 Hrs. from the start of the test). I monitor the temps with CoreTemp, and the consistent maximum reading is about 53-54 degs C, so I'm not sure if the cause of the shutdown is heat-related.

I've ruled out the PSU as a potential cause as well; I've run these tests on both a Corsair HX620W and a ThermalTake ToughPower 750W, and CPU-Z and BIOS-level voltage monitoring reveal nothing alarming.

The aforementioned RAM testing was also done with MemTest86+ meticulously, and I'm at the maximum tested stable RAM frequency with the appropriate amount of VDimm (2.75V; higher or lower throws up MemTest86+ errors). Maximum tested RAM frequency was 248MHz.

What may be causing the shutdowns?

(As an aside, running lower-priority -- i.e., 1-5 -- Orthos testing, the machine passes the 24-hour limit easily with no errors.)

Thanks so much for your thoughts!

This is where some will defer Noubourne...many feel Blend stress the system components overall more vs small fft(cpu test), large fft(cpu and little ram)
blend seems to have more of an effect on mutliple components..cpu, ram, mobo

As I said above..I found it interesting that small fft created more heat on my e6600(large fft always did more heat on my opteron)...I also found my chipset got hotter on blend..since the cpu, ram etc was being stressed

I personally believe if you are failing blend there is something overstressed....lower the ram/cpu settings and I bet you will pass

My most stable systems have been a 15hr blend test A64 and opteron that past 15hrs of large fft....

I really do put much faith in small fft as it simply stresses cpu and not the ram....

so i run a quck test on small fft to see if I my cpu is up to task..then usually run large fft for heat, cpu, ram....but in theory(maybe my opinion only..blend should more a more overall stress test)

But nothing is perfect, there are so many reports of systems that fail prime and have no issues and ones that pass 24hr blend test and fail games and certain

the stress testing does not take into account how drivers and apps play togeher..one buggy driver or program can kill any systems stability

 

[Noubourne]

Originally posted by: nealh
the stress testing does not take into account how drivers and apps play togeher..one buggy driver or program can kill any systems stability

Yes software is still a variable.

There is a new version of Prime95 that does two threads. I don't know the difference between Orthos' implementation of doing Prime95 with 2 threads vs. the original Prime95 program, but I have always used Prime95. You might try that and see if you get the same results.

You might try shutting down all other software too, in case Orthos is somehow interacting with that? Still, the immediate shutdown is weird.

I'm not sure what to tell you about the Orthos shutdown. I've had Prime95 throw an error, but I've never had it shut down the system like that.

There is another system stressing software called OCCT, that some people are fond of. You might try to see if you have similar results with that.

Otherwise, if your rig always runs fine doing everything else, and only Orthos causes it to shut down like that, then I would say screw Orthos and probably just stick with that overclock until I found some other symptom of hardware instability.

Is it possible to run Prime95/Orthos on a boot disk and test it without the OS in the way? That might be a good test to see if it was really hardware or software.

 

[Noubourne]

Prime95 is included on Ultimate BootCD.

You might be able to put the exact version of Orthos or a dual-threaded Prime95 onto Bart's PE bootable CD

 

[txtmstrjoe]

:thumbsup:

Noubourne and NealH, thank you so much for your time, attention, and expertise. I truly appreciate it.

After analyzing my particular situation, I'm more inclined to think that software issues are triggering the Orthos shutdowns. I've systematically eliminated hardware from the equation, so the only explanations left all point to either program conflicts or driver issues.

I've also examined WinXp's System Event Viewer, and there seems to be a proliferation of graphics driver errors that didn't show up before my overclocking adventures, as well as network issues. These two subsystems are the only dodgy ones in my rig, so if I had to point a finger at possible causes of system instability, those might be it.

If worse comes to worse, I can always reinstall my OS and start over. Most of my data are in separate partitions anyway.

Thanks again, guys, for your help! You both are much appreciated.