Prime 95 blend test freeze

Blend is not the correct test to run. Small FFT's is the correct test to run to determine thermal limit and stability. Ten minutes to determine the thermal threshold and 24 hours to verify stability. You should also not run any version of Prime that is newer than version 26.6. Newer versions are not optimal for thermal or stability testing as the code structure is not optimal and can create unrealistic parameters. You don't want to run AVX which newer versions do and you don't want to run Blend as it's testing memory, cpu, etc. You want to run small FFT's.

There are more detailed explanations, but suffice to say, that's the short answer.

Prime95 v26.6: http://windows-downloads-center.blogspot.com/2011/04/prime95-266.html

If it still freezes running 26.6 on small FFTs then there is a stability issue and the multiplier is too high for the amount of voltage being used or there is an actual issue with the cpu. Try increasing the voltage very slightly and restest, keeping a close eye on temperatures.

Is the CPU overclocked?

I don't know. Did you run Prime again using the version I linked to? Did the temps stay below 80 degrees Celcius for ten minutes? Were there any further errors? Run Prime again using Small FFTs and the version I lined to above. See how it does. It should be fine.

Shadrack,

Running the latest version of P95 is a good idea. The newer versions have branches of hardware optimized code that will achieve higher workload saturation on your CPU, making for a more thorough test of the CPU.

Running small FFT's, in-place large FFT's, and Blends each have their place. There is no "right" or "wrong" one to use in an overall sense as each should be used to help test more specific things.

If you crashed on a blend test, the error could have come from the CPU itself, the memory controller/L3 (CPU-NB), the memory, or anywhere in-between.



What CPU heatink are you using?

What motherboard are you using?

What PSU are you using?

What case are you using? Any extra fans?

7 hours of prime95 before a problem on a system that is configured at "stock" speeds suggests the possibility that a component was on the edge of passing grade when it was sold to you, and over the course of those 7 hours of heat and abuse, fatigued out of spec so to speak. I'm concerned that future instability even in normal use may now be a concern. It's also possible you'll never have a hardware related instability incident. Ultimately, you have to decide how important tracking down the source of that freeze is to you. (you have the option of not worrying about it!)

I'm just trying to "poke and prod" here a bit to feel this out. Looking for any obvious signs of bad system config that could be to blame. I'm not familiar with EVO LABs as a PSU brand, doesn't mean it's a bad PSU but usually if I haven't heard of it, then it's a rebrand of a rebrand of a rebrand sort of thing... In rare cases, a PSU can be the cause of instability, and an "off-brand" PSU like this does leave me uneasy feeling. Granted, the power level demanded of a Kaveri chip running P95 without any other stress tests going on (like on a GPU) is actually a pretty low load for most ATX Desktop PSU's.

Good job on using AMD OverDrive for thermal margin readings (most people don't know about that). Sounds like everything is fine there... For these long stress tests, the stock cooler is often the best thing for budget motherboards, as it provides much needed cooling to the VRM mosfets. Obviously the stock cooler is giving you plenty of thermal headroom on the CPU so that's going to be fine in your application. Inexpensive boards have bare-minimum VRM designs that aren't really designed for continuous operation at peak loads... VRMs can be the source of instability, but it will usually show up a lot sooner than 7 hours in, so I don't think that's your issue. Being a Kavari CPU, you're also below the worst case scenario for the socket (which would be an A10-6800K, which easily draws ~25% more power or more under these conditions).


The fact that the system actually froze up, rather than "dropping" a worker, suggests to me that the instability is more likely in the memory side of the equation. I forgot to task, what memory are you running?

I'd be tempted to run memtest86+... you'll get the most thorough results by making a memtest86+ usb boot drive (google it). If memtest86+ throws errors (let it run for at least a few hours), then your RAM might need to take a trip back to where it came from... alternatively, you could probably fine tune out the instability with a manual clock/timing/voltage setting.

I differ with you on the thermal cap reached in ten limits as it isn't true. I have run Prime95 on dozens of my AMD builds and on my 8320 with a Sabertooth 990FX R2 and my 6300 with my GA-990FXA-UD3 R4 both running at 4.7 GHz and the heat really starts to kick in about a half an hour in. The Thermal Margin will fall from around 68c on both cpu's to about 20c to 25c then stabilize. After a half hour to 45 minutes. After 10 minutes that margin is still around 40c. I use similiar cases and fans. The 8 core has a Noctua DH-N14. The 6 core has a Thermalright Ultra-120 eXtreme.

If you have a decent board, a mid range cooler and some good air flow in your case you should learn how to a do at least a few hundred MHz overclock. AFTER you iron out your stability problem.

At ten minutes, you will have reached the max temp likely to EVER be encountered during actual use. The temps experienced beyond ten minutes do not depict reality during normal computing, even if gaming or running extremely demanding applications in all but the most extreme circumstances. As well, once you do your stability run of 24hrs you should be periodically monitoring your sensors and Prime itself, and in doing so have the opportunity to see if that has become an issue.

And, I'm not certain that behavior is accurate using non-AVX versions of Prime, which is what I suggested he use. I guess there are bound to be differing views in all manner of test procedures, and mine is simply one of them. Doesn't mean others are invalid or less efficient.

mdocod is a legend in tech forums. Your thermal sensor isn't broken. The only way to know if an apu is within safe temp ranges is by calculating the thermal margin. Which AMD Overdrive does. You should have at least a couple intake fans and at least one exhaust fan. The cpu fan needs cool air within your case to keep your cpu cool. You have to bring cool air in an expel the hot air via case fans.

Memtest won't always throw errors in every situation. It's a good idea to run it, but if all modules pass, there can still be memory issues. We've seen many instances of modules that passed testing but were in fact still at fault and once replaced, no further issues presented. But it's still a good idea to run when memory is suspect as those cases are the exception and not the rule. Just don't assume because the modules pass, there can't still be a memory related issue.

Running Prime95 for 10 minutes proves nothing. It is hardly a test of stability. Most people want to verify rock solid stability. Especially overclockers. Maybe not you but some of the best overclockers in the various forums online swear by it. I go by evidence I verify myself. Not by anecdotal bs. Ten minutes of Prime95 is a bogus test as it hardly proves a worker won't drop out under duress. Two hours of running Prime95 and doing some encoding, playing some tunes and surfing the web at the same time is a good test of stability. That is why I never have issues running my 6300, 8320 at 1200 MHz over stock all cores.

Clearly, you can't read. I didn't say ANYTHING about running prime for ten minutes to verify stability. I said 10 minutes for thermals and 24 hours for stability. In two places I said that. Here is one. You should fully read the posts before chewing ass.

As well, once you do your stability run of 24hrs you should be periodically monitoring your sensors and Prime itself, and in doing so have the opportunity to see if that has become an issue.

Click to expand...

Since your system should have no stability issues at default settings, and longer runs are mainly for determining the stability of an overclocked configuration, if you have not had a thermal issue by now, you are not likely to have one by letting it run further. I'd say it was fine at this point. It was probably fine at 10 minutes or an hour at most.

I'd call it good.

Considering it's freezing while gaming, I'm not sure I'd "call that good."

a 7 hour stint in P95 before a freeze, but freezing in games tells me to look for the instability elsewhere...

The next thing in line, is actually the system drive. Failing system drives will trigger sudden freezes and other problems. The freeze in the middle of your P95 test may have been unrelated to the P95 workload entirely. Routine drive access may have caused the freeze. Corrupted files from a failing system drive can also cause this.

Were games freezing before you did that initial 7 hours P95 test? Or was the computer freezing in general before that test?

I'm not getting a clear answer here....

I am trying to determine if the games were freezing before you ran prime95 for 7 hours. Or only since.