Prime95 Fatal error

Jegethy

Jegethy

Jegethy

Suspended
Joined
15 Dec 2009
Posts
1,493
Location
Bristol
I overclocked my cpu to 4.5Ghz and has it stable for 2 days however I never ran Prime95 for a reasonable amount of time (around 30mins).

I ran it today, and after 2 hours on the Blend setting, I had this error:

[Sun Aug 07 11:17:29 2011]
FATAL ERROR: Rounding was 0.4998091459, expected less than 0.4
Hardware failure detected, consult stress.txt file.

The only thing I have done to OC is change the multiplier and left the vcore on auto. Also, my ram is rated at 1600MHz but is only running at 1333MHz.

I've had a look around some forums and apparently the rounding error is caused by RAM timings, however mine are all at stock speeds.

I have been running the Small FFT test to rule out my cpu and I have no errors after 4 hours.
 
Vcore more than likely mate.. might up your RAM volts a notch or two but most rounding errors i've come across are cpu voltage related, possibly heat, what are your load temps?
 
Justintime said:
Vcore more than likely mate.. might up your RAM volts a notch or two but most rounding errors i've come across are cpu voltage related, possibly heat, what are your load temps?
Click to expand...

Fairly low. Haven't hit 60'c yet. Max is 59, rest are 56-58

Oh, and my vcore is on 1.320
 
What ram do you have. If its 1.65 v ram and you have not set it to that it will be at 1.50v

as said its probably just needs more vcore. You might even need more than 1.33v
 
That's just greedy give me two sticks lol.

It says 1.50-1.65 so it could be that because you have so much that you need 1.65v

try upping it to that and run the chip on same volts
 
To save time use intel burn test instead if it passes 50 runs of that it's pretty much stable.
 
Not really, prime detects a lot of instabilities that IBT dosen't, its more refined imo especially the latest version, very aggressive on blend and somehow roots out SB instability more efficiently than say Bloomfield, no idea why. Most IBT errors i've seen are heat or really voltage starved overclocks.
 
Well, so far so good. I've been running Prime on the Blend option for just over 2 hours now with no errors. Earlier I had an error just before the 2 hour mark.

I will run it over night tonight and hope for the best.
 
Justintime said:
Not really, prime detects a lot of instabilities that IBT dosen't, its more refined imo especially the latest version, very aggressive on blend and somehow roots out SB instability more efficiently than say Bloomfield, no idea why. Most IBT errors i've seen are heat or really voltage starved overclocks.
Click to expand...

show me an IBT stable overclock that fails prime ?
 
Okay, so I left it overnight and after 30mins I got the same blending error. I bumped up the ram voltage to 1.65v and then ran Prime again.

3 hours later, no errors. I also did 10 runs of IntelBurnTest which completed successfully.

Will run Prime overnight again tonight.
 
Like C64 I'm also looking for rigs which will pass IBT but fail Prime, as I believe it's impossible.

Prime you can just click go and leave it, but IBT needs a little more care. You must use sufficient memory to get the GFLOPS column within say 80% of the max, which for Sandybridge is 32 times the clock speed (in GHz) for IBT v2.51 on win 7.

E.g. at 4.5 GHZ the max will be 32*4.5=144 GFLOPS. 80% is 115 GFLOPS. So need to use enough memory go get above this.

On "standard" you won't get anywhere near that.
 
One thing that I have been wondering about is whether operations that require more cycles, for example, the evaluation of trigonometric functions, used in prime95, are more likely to fail on unstable overclocks. Linpack in IBT uses only additions and multiplications which could make it a bit more tolerant.

On Intel x86 processors, floating point addition and subtraction require 6 clock cycles, multiplication requires 8 clock cycles, and division 30-44 clock cycles. But sines and cosines take between 180 and 280 clock cycles. This is slightly outdated info, but the ratios should be more or less correct even for the newest processors.
 
Last edited:
joeyjojo said:
Like C64 I'm also looking for rigs which will pass IBT but fail Prime, as I believe it's impossible.

.
Click to expand...

I've had enough rigs that pass 100+ runs of IBT or LINX and fail prime after a few hours to know thats its more than possible. I hate to bring up the fact that i've probably clocked more chips than the average guy here but it is a fact. I know how to set up IBT to run properly and have been using it since it had no interface and you had to check runs manually.
I've found moreso with the last few versions of prime that it has improved a lot and is rather brutal on rigs that are on the edge.
 
Left it again last night and had the same error. For some reason when I woke up, the cpu voltage was set to 1.420v which was on Auto. I went back into the bois and set the ram volatage back to auto, and changed the cpu voltage to 1.35v

I then ran Prime95 for 7 hours without an error. I'm going to run it overnight tonight and hopefully, with any luck, it won't give me an error. Only ever seems to do it when I'm not at the computer :rolleyes:
 
You must log in or register to reply here.
Back
Top Bottom