Prime95, V28.10 Build1 x64, Lockup Stopping workers
 

Hi,

Hate to post here, and I hope this is the right forum. Google seems to be failing me...

Prime 95 runs without errors, until I "Stop" all workers. Just finished a 44 Hour Stress test with Small FFT's, showing no failures in any 8 of the workers. As soon as I hit the "Stop" button to stop all threads, the system locked up hard. No BSOD, No warnings, just complete lockup. I've had this happen a number of times now.

Running Win10, 64bit, 16Gb,Ram, FX-9590, mild OC at 4800Mhz, Turbo turned off. LLC adjusted to keep core voltage at same level as no load. Temps were 20C below the thermal Limit. (49C, Thermal limit 70C)

During the run, I was able to open web pages, format drives, run system monitors, and use the PC as normal. No hint of anything unusual.

I've tried to google this, and am having trouble finding info on lockups stopping the workers. If any of you have any ideas what may be going on, please post some links or give me a few hints. I'm inclined to believe the build stable. However I'm not 100% sure it isn't hardware related.

If I can get this to repeat reliably, I may put a good scope on the board to check core voltage, etc. to see what the transients are on load step changes.

Also, I'm very sure Prime95 is the best stress test out there! :) So, It could very well be hardware related, I'm just not sure where to look... :)

Thanks,
Dan

If you run the CPU at stock clocks, does it still happen?

I had something similar happen on a Haswell overclocked system I built. I surmised that the CPU had problems backing down from the 0.1V voltage increase from prime95's use of AVX instructions and/or the voltage drop due to power-saving kicking in.

IIRC, either the problem went away when I disabled C states or I just ignored the problem since I simply run prime95 24/7.

Stability
 

Hi Mark,

So I believe it is stable at stock SPEEDS, but let me qualify. It is not stable at all stock settings, as the vdroop is a bit too much with default settings for the board/chip. A small amount of LLC was needed. Most of the time stopping all workers causes no problems. However it happens once in a while even at stock speeds (4.7Ghz).

I should clarify, the lockup doesn't happen all the time at 4.8Ghz, just once in a while stopping workers. Other than stopping workers, it's hasn't locked up.


Prime95,

Yeah, I'm guessing the VRM transients may be the root of the problem. Just 'feels' like a load dump issue on the VRM. I may put a scope on the VRM output to see what the transients are like. This system is a Frankenstein, and the FX-9590 labeled chips from AMD are a bit dubious anyway...

Might pull the clock back to 4.5 or 4.3 and try it there. If I can get it unconditionally stable I'll use it as a number crunching box or similar...


Thanks!
Dan

Have you tried a different power supply, too?

[QUOTE=Mark Rose;458798]Have you tried a different power supply, too?[/QUOTE]

Yes.

Stock supply in the case was 350W. That dropped to 11.6V under load at the CPU VRM supply pins. Plopped a 750W supply in there to see if it helped. It did not. That drops to 11.95 under load at the CPU VRM supply pins.

No high end video (Just a K620 Quadro), one SSD, and two 1Gb platter drives is all that's in the system.


Dan

11.6 V is in spec, but not great.

That being said, I don't know what else to suggest.

[QUOTE=Mark Rose;458807]11.6 V is in spec, but not great.

That being said, I don't know what else to suggest.[/QUOTE]

Yeah, it's a little weird. I didn't really want to bother everyone on the forum, but wanted to make sure it wasn't a common P95 issue. At this point it sounds like it's not Prime95. Therefore it's probably a safe assumption it's a hardware issue...

I'm leaning towards a VRM issue on the main board. In laws were in town the last few days. So maybe this evening I'll throw my scope on the Vcore rail to see how it responds to transients. If I can find anything definitive, I'll be sure to post it.

Thanks!
Dan

a AMD FX-9590 is a frankenstein monster, stock it is 220W and 4,7GHz. Many motherboards don't have the proper voltage regulation to cope with that kind of power...
Did you try disabling C-states in the BIOS?

Edit:
This thread belongs in the Q&A forum.

[QUOTE=dkemppai;458837]So maybe this evening I'll throw my scope on the Vcore rail to see how it responds to transients. If I can find anything definitive, I'll be sure to post it.[/QUOTE]

You have a scope? Cool.

I haven't played with one of those for many years. But a partner of mine had two, one of which was a multi-probe digital scope with a trigger and memory. It helped us figure out a subtle problem with an analogue circuit he had designed.

[QUOTE=VictordeHolland;458860]a AMD FX-9590 is a frankenstein monster, stock it is 220W and 4,7GHz. Many motherboards don't have the proper voltage regulation to cope with that kind of power...
Did you try disabling C-states in the BIOS?

Edit:
This thread belongs in the Q&A forum.[/QUOTE]

All of the power saving features have been disabled in the bios.

Just tried to scope the Vcore voltage, and one of the inductors for the VRM had a bad solder joint on it! It looks like it never flowed, or flowed poorly and melted later. More testing to be done now that it's reflowed.

If this thread belongs elsewhere, and you have the ability please move it to it's correct location.

Thanks,
Dan

[QUOTE=dkemppai;458881]Just tried to scope the Vcore voltage, and one of the inductors for the VRM had a bad solder joint on it! It looks like it never flowed, or flowed poorly and melted later. More testing to be done now that it's reflowed.[/QUOTE]

Just to share, I find that most people don't know how to solder correctly. It doesn't matter if it's a circuit board and components, or a copper pipe.

Most people melt the solder with the iron, and expect it to affix to the metals to be connected.

Instead what should be done is to carefully heat the metals to be affixed, and let the solder melt.

Long ago we used to hold "soldering parties" for small production runs, experienced guys coming around to solder for beer and pizza. One guy was so good that he could solder in a 40 pin strip in about two seconds.

He would run the soldering iron between the PCB and the component, and then closely follow with the solder. Then he would take a bite of his pizza, take a swig of beer, and then repeat.

[QUOTE=chalsall;458943]Just to share, I find that most people don't know how to solder correctly. It doesn't matter if it's a circuit board and components, or a copper pipe.
[/QUOTE]

In this case, a little liquid RMA flux, good temp on a very large tip was needed. Flow took a while, so there are probably large copper pours and this is probably built on heavy copper also. Once it flowed I added a little more solder, and it looks as good as the rest of the joints.

So, it's been running steady now, at 4.9Ghz with less voltage than needed previously at 4.8Ghz. So far start/stop Prime95 has been without trouble since the solder joint.

More testing to see how it holds out!

Thanks,
Dan

[QUOTE=dkemppai;459011]In this case, a little liquid RMA flux, good temp on a very large tip was needed.[/QUOTE]

Yeah, flux is important; helps with the "wetting". Thank you for reminding me of that.

Edit: I do find it a little odd that LaurV hasn't yet spoken on this topic. He regularly works on surface mount stuff. :razz:

[QUOTE=chalsall;459156]...He regularly works on surface mount stuff. :razz:[/QUOTE]

It's been solid so far. Lost an old platter drive in it today, but that was an old junker just for scratch space...

Yeah, my day job is designing and build of analog, RF, digital, and power circuits. Part of my job description is to instruct people on soldering technique, and inspecting their work. We use a lot of SMT components, although we still try to avoid flip chip type and BGA packages as inspection is too difficult. Some of the bigger boards we've done are close over 1000 components, with a good mix of multi Ghz stuff, and associated passives. We're just starting to play with some 2 to 3 Gbit/second serial link stuff too. There's probably close to 100 different types of solders, some better for specialty applications than others...

I still find it hard to believe that this board had a cold solder on it. Although, that may explain why I picked the whole thing up so cheap. Wasn't really shopping for a FX-9590 system, but it was offered at a price I couldn't refuse.

Just goes to show, Prime95 stress test can point out problems that 99.99% of the time will never show up... :)

Thanks,
Dan