SOLVED Prime95 Worker #5 has stopped

[Snivory]

Hello, I'm new to these forums, I'm not too sure if I'm posting in the right section... anyway I was wondering, what do I do if a worker stops in Prime95? Because with IBT, I was stable with 20 runs. My current OC and hardware are:

Gigabyte 990fxa-ud3
1090T BE
Multiplier: 18.5x
FSB: 228mhz
Vcore: 1.475v (bumps to 1.54 under load, LLC is set to Regular, don't know why this happens)
NB & HT Link Frequency: 2280mhz


So, should I revert to my stable Prime95 settings? Because, to be honest, I don't know what they are. I've been using IBT because it's much faster...

Anyway, here's my Prime95 log.
[Tue Oct 09 08:21:43 2012]
Self-test 640K passed!
Self-test 640K passed!
Self-test 640K passed!
Self-test 640K passed!
Self-test 640K passed!
Self-test 640K passed!
[Tue Oct 09 08:37:22 2012]
Self-test 8K passed!
Self-test 8K passed!
Self-test 8K passed!
Self-test 8K passed!
Self-test 8K passed!
Self-test 8K passed!
[Tue Oct 09 08:45:01 2012]
FATAL ERROR: Rounding was 0.4990234375, expected less than 0.4
Hardware failure detected, consult stress.txt file.

23 minutes and worker #5 stops working. So I consulted stress.txt and it led me here.

Also, while I'm here, would anyone know to increase CPU NB voltage on Gigabyte 990-fxa-ud3? In BIOS, there's NB voltage, which is for the mobo NB, right? And then there's CPU NB VID Voltage, which I'm not too sure is the same as the CPU NB voltage...

Any response is appreciated, thanks in advance.

 

[keny]

Hi! And welcome to the forum, the usual cause for a worker stopping is a vcore problem but at 1.54v under load I would say maybe not. I would resort you're HT back to 2000 and set you're LLC to something else, I have mine set to ultra high, and voltage hardly moves under load on my 990fx ud3, you usually have to try multiple settings on LLC to see which one works best for you. As for the CPU NB setting, you are correct in assuming that CPU NB vid us the correct setting for the CPU NB voltage. If you could take a screen of cpu z main spd and memory tab and upload it, it would also help try and diagnose any problems you are experiencing.

Mark

 

[Zes]

I'll give you the same recommendations a friend gave me when something similar to this happened and it worked (actually, I've had more than one worker crash)

- Get the latest version of Prime95
- If workers keep crashing, try switching LLC to "Enabled" if possible and if you haven't done that yet. Make sure to re-test the voltages just in case.
- You can also try bumping the core voltage in baby-steps and try P95 again.

Looking at the logs, I would suggest running a 30min run for each test or an hour if possible. I'm at work atm so I can't be much more specific, I'll get back into this when I get home.

Good luck my friend! Hope some of this works for you!

 

[EarthDog]

*LLC IS enabled (he has it on regular and has vrise)
*Voltage is already high for those clocks...

 

[BeepBeep2]

This is the vrise I've been preaching about.
keny, you have a UD3 or UD5? Is it UD3 1.2 or higher? 1.1 has this vrise I know confirmed, didn't expect GB to do anything about it further.

REGARDLESS, the rounding error is related to CPU-NB/IMC or memory. Increase CPU-NB voltage .025v, if it helps great, if not, then drop it back down and increase memory voltage .025v to see if that helps.

On all Gigabyte AMD boards, CPU-NB voltage is listed as "CPU/NB VID Voltage"

 

[keny]

I'm lucky mate, I got a ud3 1.2 but I know that the 1.1 has a frequency fluctuation problem, sort of like CnQ but throttles like hell, my ud3 did it at first but a bios update sorted it out but I don't think it is rectifiable on the 1.1 ver

Unless his memory has been backed down a notch it may be overclocked too much at that ref clock

 

[Zes]

@EarthDog :

My bad! Deeply sorry! I completely missed a few things he mentioned. Thanks for pointing it out though.

I'd go with BeepBeep2's suggestion.

Out of curiosity, what are the temps while under load?

 

[Snivory]

I would resort you're HT back to 2000 and set you're LLC to something else, I have mine set to ultra high, and voltage hardly moves under load on my 990fx ud3, you usually have to try multiple settings on LLC to see which one works best for you.

Sorry for taking so long in replying, I left prime95 overnight with NB and HT @ 1996mhz through the multiplier, not FSB. Was I supposed to do it through FSB? Because worker #5 took longer to stop, around an hour before I got the error.

REGARDLESS, the rounding error is related to CPU-NB/IMC or memory. Increase CPU-NB voltage .025v, if it helps great, if not, then drop it back down and increase memory voltage .025v to see if that helps.

Bumping CPU-NB seemed to help. I just did it right now, and my PC booted faster. Also everything seems more stable on the desktop, but I've yet to run any stability tests. I'll be doing that right now.

Thanks for the replies, guys.

 

[keny]

Yeah you did it right by using the multi, but you would have been ok leaving the CPU NB at 2280 and bumping the CPU NB voltage like beep suggested, what is you're memory running at? As I think it may be you're memory that's stressed and causing the failure, you can find our by looking at the CPU z memory tab

 

[Snivory]

Sorry keny, that slipped my mind..





So far, things have been stable under 10 runs of IBT with NB @ 2280mhz/1.2v , so I kinda bumped the FSB, lowered the multiplier, and that's how I got the clean 4.2, which has been stable under 10 runs of IBT. I've yet to run prime. Also, is it possible to rule out the RAM through small FFT prime test?

The RAM timings are weird, because at stock, I had them at 9-9-9-24. I've never touched RAM settings before, it's completely uncharted territory for me.

 

[BeepBeep2]

To rule out ram, use Prime95 Blend.

If you get the rounding error again, it will be either CPU-NB or DRAM, so either lower CPU-NB clock, or add more voltage 0.025v at a time, or work with DRAM the same way.

I wouldn't recommend overclocking both at once, since you won't know which one is failing when you get an error.

 

[Snivory]

Huh. Prime95 says my CPU is AMD K10. Is this a cause for concern?

Edit: I know Phenom II uses K10 architecture, but Prime won't load more than 2 workers.

 

[keny]

It is a k10

 

[keny]

Can you do the memory tab also please

 

[Snivory]

Yeah, I know it's a K10, I was just being retarded, haha. And, I'm pretty sure I uploaded the memory tab up there. At least, I think so. I mean, I can see it with my browser. In any case...

http://imageshack.us/photo/my-images/26/cpuzzr.png/
http://imageshack.us/photo/my-images/39/cpuzram.png/

 

[keny]

Sorry, my turn to be retarded !!!! I mean the SPD tab

 

[Snivory]

 

[BeepBeep2]

Prime won't load more than two workers?
Can you post a screenshot please? Thanks.

 

[keny]

Ok, try this for now, put you're ht no higher than 2000, drop you're ram devider to 665 or whatever jedec#4 says as I can't see it very well as I'm using the mobile site, also set the timings to jedec #4 (9-9-9-25-34) and bump the ram voltage to 1.55v , if it fails after this then we know its not you're ram. You have not mentioned what temps you are getting under load, you can monitor this using HWmonitor and you're temps want to stay under 55c under load

 

[Snivory]

Prime won't load more than two workers?
Can you post a screenshot please? Thanks.

Yeah, I thought it was weird at first, but when I began the torture test, it loaded all 6 six workers, and all cores shoot up to 100%, so I'm pretty sure it's all good.

30 mins and so far, no workers have stopped.

You have not mentioned what temps you are getting under load, you can monitor this using HWmonitor and you're temps want to stay under 55c under load

Oh yeah, that's right, my temps are 51 C under load. vCore manages to stay at 1.54. I figured those temps were good, so I kinda neglected to mention them.