EDITED - Question Re: Prime 95 Error Running Stock settings

[Herkulese]

I ran Prime 95 with 4 workers, and at priority 10.
I had it set for small FFT’s, which as I understand it, stresses the CPU the most.

The system was running at completely stock speeds, and with fail safe BIOS settings.

Board: Gigabyte 970A-UD3 (2nd Hand)
CPU: Phenom II X4 965 (2nd Hand)
RAM: G.Skill Ares 8gb (2x4gb) 1600 (New)
PSU: SeaSonic G Series Modular 550 Gold + (New)

I started it at 9:00 pm Tuesday night, and let it run until 5:00pm last evening, for a total of 20 hours.
My temps never got above 51 degrees, and held more down at 48ish area.

RESULTS: The first three cores were still running, error free, but at Hour 8.5 the forth core had one error, and stopped working.

QUESTION: How concerned should I be about this, and what, if anything should I try to do about it?

Thanks

 

[VirtualLarry]

Yes, you should be concerned. Either your motherboard's power-supply circuitry (VRMs, caps) are going out, or your CPU is damaged. Were you overclocking it before (overvolting)?

Edit: Could be RAM too, I suppose.

 

[DrMrLordX]

If it's small FFTs, it's not likely the RAM. I'd expect that more from blend.

If you are having a core go down during testing, then it means that you are unstable, though not catastrophically so.

One option is to improve your cooling and try again, though I can't promise that will help you any. Is there any way for you to blow some extra air at your VRMs and see if that fixes the problem? It may be that you are seeing voltage ripple from the PSU or board operating out of spec somehow.

Another option is to up vcore and try again. This may exacerbate the problem in the long term, but in the short term, it may be your only solution. Or you can just shut down core #4 in the BIOS and run with a tri-core.

 

[mindbomb]

ram or hypertransport frequency can cause failures even when core is unchanged. They can be from bios mistakes in automatically setting the right frequency. Try manually setting timings, ram freq, and htt frequency. These errors are unlikely in small fft though. Motherboard and power supply health is also suspect.

 

[DrMrLordX]

One thing I'd like to know: is it always core #4 undergoing failure? If so, I doubt that it is a power delivery issue, unless AMD set up the power plane to deliver uneven voltage to cores in the event of marginal voltage delivery failure.

In other words, if it were the PSU or board/VRMs, I'd expect to see semi-random failure of cores, or failure of multiple cores. It looks like only one is consistently unstable.

 

[Herkulese]

Yes, you should be concerned. Either your motherboard's power-supply circuitry (VRMs, caps) are going out, or your CPU is damaged. Were you overclocking it before (overvolting)?

Edit: Could be RAM too, I suppose.

No I have not overclocked at all on this new buildl. I was running a standard settings Prime 95 test, just to ensure that it was good to go at stock, before considering an over clock.

I invested in a SeaSonic G series 550 Gold+ power supply, so I don’t believe the issue would lie there.

I bought this motherboard and CPU used from a forum member, who had only over clocked it once just to test it out, but ran it mostly at stock.

I guess I need to run memtest just to see if the memory is solid, before making any other decisions.

Any other suggestions?

 

[Herkulese]

One thing I'd like to know: is it always core #4 undergoing failure? If so, I doubt that it is a power delivery issue, unless AMD set up the power plane to deliver uneven voltage to cores in the event of marginal voltage delivery failure.

In other words, if it were the PSU or board/VRMs, I'd expect to see semi-random failure of cores, or failure of multiple cores. It looks like only one is consistently unstable.

That is an excelent point. Maybe I should run the test again, and see if core 4fails again, or if another one does.

 

[DrMrLordX]

Sounds like a plan. Running Memtest86+ and the like is also a good idea, just in case.

 

[Herkulese]

Sounds like a plan. Running Memtest86+ and the like is also a good idea, just in case.

I will run memtest first, then see where things stand, then go back and run Prime 95 to see if and/or which core fails.

How long should memtest be run, or is it test specific?
It has been a long time since I have used it.

Thanks

 

[DrMrLordX]

The last time I saw any recommendations for Memtest86+, it was noted that the most rigorous testing occurs in tests 1-5 or 1-6. You can set it up to exclude the later tests (if you like) and let it loop for maybe 4+ hours. I usually let it go when I go to sleep and then stop it when I need the computer later, depending on how badly I need the machine.

Of course, I also usually let it run the other tests as well, just for the heck of it.

 

[Herkulese]

The last time I saw any recommendations for Memtest86+, it was noted that the most rigorous testing occurs in tests 1-5 or 1-6. You can set it up to exclude the later tests (if you like) and let it loop for maybe 4+ hours. I usually let it go when I go to sleep and then stop it when I need the computer later, depending on how badly I need the machine.

Of course, I also usually let it run the other tests as well, just for the heck of it.

OK, I now have my MB BIOS set to Optimized Defaults, and ran the latest version of Memtest86 from a boot disk. I rean it from 8:30 last night to 6:00 this morning and it survived 5 passes with no errors. I am fairly confident that my new G.Skill Ares Orange ram is good and stable.

I will now go back and run Prime 95 again.

Now, I have to admit a very dumb thing that I did when I first ran the test. It had been so long since I built a new system, and even since I had hooked up my current one, that I totally spaced the need to plug it into a good Surge Protector. In my hast, and given the position of the unit when I went to first fire it up, I stupidly plugged it directly into the wall. Now my plugs are grounded, but just the same, it could have seen voltage spikes, which I am in hopes caused the error that I saw, without causing any permanent damage.

Needless to say, I immediately switched it to my surge protector, and I am glad to see that my ram doesn’t seem effected.

I will post back how it goes after the next Prime95 test.

 

[DrMrLordX]

Let us know how the prime95 testing goes. It would be premature to point fingers at your home's wiring, but then again, I have lived in at least one place with some truly awful wiring that was pretty marginal when it came to powering computers.

 

[Herkulese]

Let us know how the prime95 testing goes. It would be premature to point fingers at your home's wiring, but then again, I have lived in at least one place with some truly awful wiring that was pretty marginal when it came to powering computers.

I am not really questioning my homes wiring, but rather the inconsistency of the voltage being fed into it. The incoming voltage can spike, and that is largely what the Surge Protector is for. To protect against those voltage surges that come from the grid.

I ran Prime95 last night, but forgot to reset so that my PC would stay running. It went to sleep after 12 tests.

I will post back when I have done another proper test.

 

[westom]

I am not really questioning my homes wiring, but rather the inconsistency of the voltage being fed into it. The incoming voltage can spike, and that is largely what the Surge Protector is for.

Actually no. That is what a PSU is for.

Voltage can vary so much that incandescent bulbs dim to 40% intensity. Even that is ideal voltage to a computer due to its power supply. If a PSU is working properly, voltages are just fine. But normal is for PSUs from any manufacturer to be defecxtive. A defective PSU can boot and run a computer almost constantly. But is still defective. Those concerns about power are eliminated and justified immediately using a meter (and some instructions). Without hard numbers, everything about power is only wild speculation. But again, a completely defective PSU can boot and run a computer most of the time. The resulting failure is intermittent; maybe infrequently. But the PSU is completely defective.

Same appliies to the Prime05 test. Intermittent at 50 degrees implies a defect. Repeat testing says nothing useful.

However that is what stress testing is about. Not runing tests for hours. Obtain the maximum CPU temperature from CPU specs. Then heat that CPU to that temperature. Only then does a stress test exist. Then tests such as Prime95 and Memtst report something useful. That heast causes no hardware damage. Heat is a diagnostic tool. If hardware is defective, then hardware does intemittent or constant software crashes at elevated temperatures. And that defect will get worse with age. Use temperature to identify timing or threshold defects in a CPU that otherwise works fine are lower temperatures.

Two techniques to first identify the defect. Fixing it comes later.

 

[DrMrLordX]

Things like a good PSU and a UPS can go a long way towards cleaning up "dirty power". That being said, if your wall socket isn't delivering consistent voltage, you may wind up with problems. I have lived in an old house (it was actually a house divided into apartment units) with bad wiring that (apparently) produced all kinds of undesirable phenomena when powering electrical appliances. Was it really the wiring? That's what everyone said it was. Maybe it was the local substation? Maybe it was something inbetween, who knows? All I know is, even with a fairly nice surge protector (NOT a UPS), I was pretty sure that I was getting dirty power out of the socket. We did get a lot of bulb dimming or (in some cases) surging, problems with simple audio equipment (boom boxes) behaving badly, and some unexpected arcing from my old PC speakers (it was usually the subwoofer that did it). Things probably would have gone better for me with a UPS, but oh well. I didn't know any better back then.

 

[westom]

That's what everyone said it was. Maybe it was the local substation? Maybe it was something inbetween, who knows? All I know is, even with a fairly nice surge protector (NOT a UPS), I was pretty sure that I was getting dirty power out of the socket.

Unfortunately, most recommendations come from "everyone said it was" reasoning. Fewer recommendations, based in reality, say so with numbers. Are bulbs dimming to 40% intensity? If not. voltage is sufficient for a computer. Move on to other suspects.

Is bulb intensity changing? Then either the AC power utility has defects or household wires are defective. In most cases, this is a trivial problem. However in rare cases, this is a major human safety issue. Don't foolishly cure symptoms with a UPS or some other magic box. Find and fix what might be a seriouis problem. Unfortunatley many instead will recommend a UPS only because others said same.

UPS addresses one anomaly. It provides temporary and 'dirty' power during blackouts. To protect data (not hardware). Too many recommend a UPS to eliminate anomalies that a UPS would not cure. Or to even 'cure' a defective PSU. No numbers in a recommendation quickly separates a useful one from hearsay.

An example. Read nuimbers provided with each 120 volt surge protector. Its let-through voltage is 330 volts. That means it does nothing until 120 volts rises well above 330 volts. How often is your AC voltage rising that high? Hourly? Then you are probably replacing clocks, radio, refrigerator, dimmer switch, and CFL bulbs daily. Too many who make recommendations ignore those damning numbers.

Is that PSU defective? Even a brand name says nothing useful. Answer requires some instructions with a meter. Otherwise, only "everyone says it was" reasoning exists.

Two techniques to actually find defects were provided. Both require or provide numbers.

 

[DrMrLordX]

It's not too hard to eliminate the PSU as a source of trouble. Plug it in somewhere else (that is, some other building, preferably one serviced by a different substation) and see if the problems persist. Or, try a different PSU that is known-good and see if that clears up trouble. Those aren't the best techniques, but they are available to those without the equipment and/or skills to test a PSU more directly.

In the case of the OP, I don't think it's a power problem at all, especially if it is the same core that continues to cause trouble. But, until he does more testing, everything is just an assumption. A UPS probably won't help him. Even a PSU change probably won't help.

I agree that a UPS does nothing to address the actual problem of dirty power coming from the socket in those circumstances where it is a problem at all, it's just that, sometimes, it's either that or a naked socket. In my case, it was the motivation of low rent that kept me in that place for as long as it did. Sometimes, entire neighborhoods can be affected by rolling brownouts and other undesirable effects with their origins far beyond the house's wiring.

There are entire countries that have power delivery problems.

So, if I knew there was someone that lived somewhere where power problems were obvious (or better yet, measured properly by someone competent), and they couldn't realistically relocate, yeah I'd understand them wanting a UPS. It is quite possible for stability problems to come from the socket or beyond, no matter what the quality of your PSU. It isn't particularly probable, though, and I'll reiterate that I don't think it's the OP's problem.

 

[Herkulese]

I am not ignoring the previous posts but I promised to post back after running another proper Prime 95 test, and I just ran Prime95 for 22 hours with no errors. The temps got up to 58c, but no issues.

 

[DrMrLordX]

Then it's fairly safe to say that you've ironed out whatever stability problems you had in the first place.

 

[Herkulese]

Then it's fairly safe to say that you've ironed out whatever stability problems you had in the first place.

Well, I am now endeavoring to OC this CPU, and got it to run Prime95 small FFT'sfor a couple of hours at 3.6ghz at 1.4v, so I went up to 3.8ghz, and had to go to 1.475v (Normal Voltage on my board) to get it stable.

I ran Prime95 for a couple of hours last night, and the temps locked in at 54c.
However this morning, when I checked on it, Core 1 had errored and stoped. The other 3 were still testing, so I shut it down. I did forget to check just how long core 1 ran before the error, allthough I know it was over 3 hours, as it was still running when I went to bed.

I did another quick test this morning.

I set the voltage to 5.0v (+.25 over Normal) and ran Prime95 for about an hour. The temps maxed out at 57c, but settled in at 36c. I did not want to let it run all day without being there to check on it, so I will run it again tonight, as soon as I get home, and if all is good when I go to bed, I will let it continue, through the night. That will be at least 14hours.

 

[DrMrLordX]

Hmm, how'd you get your temps down from 58C to 34-37C? Or is it just taking Prime95 15-20 hours to reach the 50s?

 

[Herkulese]

Hmm, how'd you get your temps down from 58C to 34-37C? Or is it just taking Prime95 15-20 hours to reach the 50s?

Oh man........... My Bad..........
The max out was 57c, not 37c
I have corrected that.

I believe that the temps are holding at just a bit lower at 1.5v than they did at 1.475 because I have the LLC set ar regular, and that seems to hold the voltage a lot closer to the set amount.

Still not bad for 1.5v, based on all that ai have read, and the fact that I am using a Noctua 120mm cooler reather that the EVO 212 140mm cooler.

Feel free to correct me if I am wrong.

 

[DrMrLordX]

It depends on which Noctua cooler you are using, I guess. I am personally not all that comfortable with temps higher than the mid 40s, but you are within the acceptable thermal envelope for the chip.

If it were me, I'd try to get it stable at 3.6 with as low vcore as I could (taking into account the actual voltage the chip is running @ load thanks to LLC) and then I'd look at NB overclocking. Even a modest NB bump will yield some improvements. 200 mhz NB is usually easy to reach.

 

[Herkulese]

It depends on which Noctua cooler you are using, I guess. I am personally not all that comfortable with temps higher than the mid 40s, but you are within the acceptable thermal envelope for the chip.

If it were me, I'd try to get it stable at 3.6 with as low vcore as I could (taking into account the actual voltage the chip is running @ load thanks to LLC) and then I'd look at NB overclocking. Even a modest NB bump will yield some improvements. 200 mhz NB is usually easy to reach.

Thanks, that is my plan.
There is a wealth of information on what are good NB, HT, etc.... settings.

One place shared these (Let me know what you think):

CPU Freq: 225
Multi: 18 (4050mhz)
NB Multi: 12 (2700mhz)
HT Multi: 9 (2025mhz)
CPU-VID Volts: 1.475
CPU-nb Volts: 1.275

 

[DrMrLordX]

Oops, meant 2400 mhz NB, not 200. Dur. I think you will be fairly lucky to hit 4 ghz with your chip, given how hot it is already getting, but if you can get there, hey, good on you. 2700 mhz NB would be a significant improvement over the stock 2 ghz.