Prime95 Torture Test Questions

I'm running the Blend torture test to test my ram.

A little background. My Cosair XMS ram is rated at 2-3-3-6. When I installed it, it default to 3-3-3-8, 2.6v DDR Vcore. Well, I did some reading, and Corsair rates their C2 series ram to go to the 2-3-3-6 timings, but it defaults to slower--to stay within spec.

So I manually set the timings, and increased the DDR vcore to 2.75v. This is specifically how Corsair tested it, and warranties it all the way to 2.9v.

Anyway, I ran Prime95 and within 15 minutes, it errored out. So I raised the voltage to 2.8v. Same problem. I was also running itunes and the visualization at the same time, in both instances. Both times, the same error occured in the SAME test group, same part.

So I went through the Prime95 Help, and found this:

So I went back, and ran Prime95 again, without iTunes, or the visualization going.

So far, the test has been going for twice as long as before, without errors. I plan to leave it for 24 hours.

So my question is:

Could other applications cause Prime95 to glitch? iTunes isn't exactly a stress testing app.

 

I have never heard of a program causing an error with prime 95,but usually when i run it it is the only thing running. You could run memtest to with and without those programs too see if the error is reproduced, if running prime 95 or memtest for 24 hours produce no errors you are good to go. usually I use prime 95 for testing the stability of the processor and use memtest for the memory and have never had any problems with and\ overclocked system or a stock system. There is this other stress test utility called OCCT I have used and it is a good utility.

LINKY

AMD Processor Support Forum -> Overclocking, Stress Testing & Benchmarking Tools

 

Thanks. I've backed it off to 2.5 CAS. 2 seems too aggressive, and I eventually received an error when running it on a single core. Corsair originally said this ram would run at 2 CAS for Intel, and 2.5 CAS for AMD machines. They recently changed it to 2 overall. So maybe there is something to that.

So, now I am running Prime95 with it at 2.5 CAS. I'm running two instances, one on each core with an 800mb memory load for each instance. Both cores are pegged at 100 percent.

We'll see. Kind of annoying really, because this is Corsair XMS ram. I might actually see if I can RMA it. Not sure if they will do it for a .5 CAS difference though. Although, I think they will, reading their support forums.

 

Did you try your mem at 2.9v?

 

No. Thats the limit Corsair warranties. Seems kind of crazy to set it to that, without me overclocking. Right now I've been running two instances Prime95 (One set to each affinity), each with an 800mb memory load for 5 hours straight with no errors. I'll let it run throughout the night.

Seems like its a 50/50 chance with this ram on AMD systems. Initially, Corsair only rated it on AMD systems for 2.5. Not sure why they changed it to 2.

I guess I could try it, but I get kind of skittish playing with voltages.

 

They probably changed chip on that model. You can see from version numbers/other details which you can look up at Corsair forum.

You should use default bios settings - or what works the best without overclocking. No presets, aggressive whatever - perhaps also not 1T, try 2T. May be even a new bios version is better.

Best to use memtest86 for the test you are after. To see if modules are ok as in having no bad sectors. Once confirmed you can start with P95. Much more picky. On my NF4 memtest86 do not have problem with 1T vs 2T at stock level or a bit above, not so with P95. .

If you know there must be a timing issue, approved memtest86 is what Corsair ask you to, you could try make a "custom" torture test. Punch in 4096 as min FFT - same as max, set ram to half of your system ram so you can use computer for lighter tasks, browsing etc. without too much swapping. Should tricker errors much faster than regular "blend". Seems to, good for when you test overclock.

You should search/read up on forum of motherboard and Corsair. Some bios just dont work well with xxx ram/chips. One of the reasons I didnt buy a DFI was it used to have problems with Ballistix which I use. Now solved with bios but didnt look good at the time. Most mb-forums will have some "what ram works and not!" thread. Takes much time to analyze own hardware so I would investigate and seek people with same hardware. But be careful about what you said with new XMS? same name though different hardware.

Unless you have one of few known issues with P95 or has discovered something new it is best you adjust until errros are gone. You have seen his stress.txt - not up for debate Anyway, I think the order of testing new computer should be memtest86 then P95, if P95 is at all necessary with default values. Any change like 1T vs 2T and it could be.

 

Well, if I go with the stock timings, I should have just saved my money and bought Corsair Value. I chose this ram, instead of faster OCZ ram due to the fact that it had been tested, at 2 CAS , on this board. Its on the Asus QVL list.

But, perhaps you are right. I spent all day yesterday researching the issue, and found no documentation of modulae changes, or anything else. I'm not sure I can research it further. I've got a post on the Corsair board, so I'll see what they say. Brand new board, A8N32-SLI Deluxe, it already has the latest BIOS when it was shipped to me.

I only changed timings. the CMD values have not been touched. I'm only using custom Prime95 values, because you cannot test dual core CPUs to their full extent without pegging both cores.

So far so good. 12 hours, no errors. I may let it go further, but truthfully, I see no need. I was going to do memtest next anyway

So...did you leave it at 1T? I'm looking for max performance, with max stability. I know the timings are also having effect on how Prime95 acts with the dual cores...if I ran Prime95 with other apps like iTunes, it crashed quickly, if not, it was ok. However when I ran two isntances, one core crashed out almost instantly.

I am still learning, but I can't see this as a good thing for stability, but I would really like my CL setting of 2. Anyone know how much of a performance hit I am taking of 2.5 vs 2?

 

Ok, I've been reading around that Prime95 is getting to be less useful as time goes on in determing hardware issues. I'm going to move on to different stress tests and memtest86+. I'll let ya know what happens

Who knows, maybe stressing both cores to 100% was extreme.

 

http://www.techreport.com/etc/2005q4/mem-latency/index.x?pg=1

Just read this. 2.5 vs. 2 isn't going to make much difference at all.

Although I'm going tos ee what I can do with the CMD.

 

12 hours is good enough, author says min 6. Feel approved. CAS 2 vs. 2.5 dont mean much, neither does 1T vs. 2T on NF4. You must try best setup of course but dont cry if only 2T works. Also a matter of cpu I think, no way 1T willl ever be stable with my 3500 cpu. Have tried 3 different sets of ram. I have MSI though so cant be sure... You need to read and understand datasheet of chipset to know more. From what Ive read and found 1T/2T depends on combination of ram, bios, cpu (mem controller) - and not as some say "budget/2T" ram vs. "1T" ram. You find out, Asus vs. Corsair should be documented very well.

No I use 2T but as said it could be due to my hw combination, MSI bios is a bit strange at times. Definitely not stable with 1T but even if it appeared to be I would not use anything that couldnt pass P95. You do seem to have an itune issue but if we forget about that 99 out of 100 not thinking highly of P95 are benchmark happy. You know overclocking... P95 is a joy/mhz killer, it just stops So there are other tools to test with - most will increase range of what seems to work. I dont want to fiddle/overclock if stability is less compared to stock computer so go by P95 blindly. Never have to reformat every other month, never a blue screen etc. CPU is not so sensitive unless on the very limit but memory and other related bios-timings are. CPU easy/forgiving, temps/volt - ram = magical bios presets/settings no one understands. Trial and error = not 100% safe. P95 is final answer for me, memtest86 not enough during overclock at least.

Ok, may be you can provoke P95 to crash by lettting it eat 100% of memory while you start up game, Photoshop - but then what is crashing? Just having Itunes running shouldnt be a problem, you could post on P95 forum perhaps. There have been issues before, very critical for such a tool - lose all credibility fast. As far as I know they have all been fixed. One I have been able to reproduce, something with 1gb and AMD cpu - 2-3 versions ago. Unfortunate but only a piece of software, not wondertool. Whole point of P95 is it cant be fooled and produce valid results so I would be happy with those 12 hours you have now. I doubt any of those P95 "issues" have made it less hysteric with testing, more like what you experience - crashes for no reason and no matter settings.

True what you read though, "you can easily have a full working overclocked/"optimized" computer which can only run P95 for 1 hour" right? An old argument. I think that is bad oc/optimization If you research who is doing what you find out why and what could happen if you disregard errors. Also note there is a difference between a normal user and one who only run benchmarks and reformat every other week because of new hardware ! Often those who use terms like "screenshot-stable" "benchmark-stable" - not what you are after. I think computer should be able to work 24 hours a day even if it cost 100mhz What is "possible" is the real issue of most anti-P95 arguments, believe me. Has been the case for years. You want mhz/fsb/timings/temps etc. to be in red area or a bit above or have a little controlled range before things gets critical? Would guess 50/50 pro and against if you was to make a poll. I hate it too but as said how you use computer is important and what decides if tool is useful or not.

 

Stressing the CPU(s) wont help find memory errors....I use Memtest to test Ram ..and Prime 95 to test core Overclocks.....If you can take 4 100% loops of mem test you should be good to go ....Also i dont see a resion why that ram wont run 2 2 2 x at the voltages your running ..Another thing, running T1 ( CPC or command per clock enabled ) will vastly boost ram speed over T2.. With my old rig CPC enabled to disabled made for a ~500MBs dif ...

 

Prime95 will test ram just as well. Stressing both cores, I was using more ram than running a single instance. 850mb per instance.

So far I've been going for almost 12 hours @ 2.5-3-3-6 1T.

 

As I understand it, small FFTs stress the CPU. Large ones stress the ram. I was doing small to large ones.

 

I'm not alone. I found others who had trouble pushing it to 2 CAS with similar motherboards, same video card, and CPU.

I think I can live with 2.5-3-3-6 1T anyway.

1T gave me killer results vs 2T. 2 CAS vs. 2.5 did not.