Rounding errors in Prime95 with new i5

Toggle sidebar Toggle sidebar
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530
I just build a new computer out of brand new parts fetched off newegg largely. When I build a new computer I always run several days worth of stability tests on the system to ensure completely 100% stable. One test I run is Prime95, I run all three different tests for 24 hours each. The first test, the "Small FFTs" test, ran for the 24 hours perfectly. Next was the "In-place FFTs" test; that ran for 2h 30m~ before thread 3 stopped with error "Rounding was 0.5 was expecting less than 0.4" (or something close to that). Did a little research, saw there was a bug with i5 CPUs that was fixed in a new BIOS version, so I updated my BIOS hoping that would fix it. Started the "Blend" test and this time it lasted for 22h~ before thread 4 had an error, "Rounding was 0.4997... was expecting less than 0.4" (or something to that effect). So I ran Memtest86+ for a full 24 hours, no errors. I have read some people say "raise the Vcore up to 1.65v" but if I set the Vcore at all it stops POSTing. Keep in mind that I have no overclocking; what could be causing those errors? How can I fix it?

Specs:
CPU: i5-4590
MOBO: AsRockZ97M Pro4
RAM: G.Skill Ripjaws x Series 8GB DDR3 1600 (PC3 12800)
HDD: WD Black 1TB
PSU: Corsair CMX series CX430M 430w
Case: SilverStone Temjin Series TJ08B-E

*I know this may not be a CPU issue, but idk where else to post it
 
Solution
CompuTronix
Do NOT run any versions of Prime95 later than 26.6. Here's why:

Core i 2nd, 3rd and 4th Generation CPU's have AVX (Advanced Vector Extension) instruction sets. Recent versions of Prime95, such as 28.5, run AVX code on the Floating Point Unit (FPU) math coprocessor, which produces extremely high temperatures. The FPU test in the stability testing utility AIDA64 shows the same results.

Prime95 v26.6 produces temperatures on 3rd and 4th Generation processors more consistent with 2nd Generation, which also have AVX instructions, but do not suffer from thermal extremes due to having a soldered Integrated Heat Spreader and a 35% larger Die.

Please download Prime95 version 26.6 -...
Sort by date Sort by votes
CompuTronix

CompuTronix

Intel Master
Moderator
Nov 13, 2006
6,374
279
38,240


Which version of Prime95?

1.65 Vcore is WAY too high! That's a quick way to kill your processor. The maximum recommended Vcore is 1.300.

1.65 refers to memory voltage.
 
Upvote 0 Downvote
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530


I should have the latest versions of prime95 the exe says file version is 28.5.1.0 and product version is 28.5.0.0

OK so then what do you recommend I try?
 
Upvote 0 Downvote
CompuTronix

CompuTronix

Intel Master
Moderator
Nov 13, 2006
6,374
279
38,240
Do NOT run any versions of Prime95 later than 26.6. Here's why:

Core i 2nd, 3rd and 4th Generation CPU's have AVX (Advanced Vector Extension) instruction sets. Recent versions of Prime95, such as 28.5, run AVX code on the Floating Point Unit (FPU) math coprocessor, which produces extremely high temperatures. The FPU test in the stability testing utility AIDA64 shows the same results.

Prime95 v26.6 produces temperatures on 3rd and 4th Generation processors more consistent with 2nd Generation, which also have AVX instructions, but do not suffer from thermal extremes due to having a soldered Integrated Heat Spreader and a 35% larger Die.

Please download Prime95 version 26.6 - http://windows-downloads-center.blogspot.com/2011/04/prime95-266.html

Use only Real Temp to measure your Core temperatures, as it was designed specifically for Intel processors: Real Temp - http://www.techpowerup.com/downloads/2089/real-temp-3-70/

Your Core temperatures will test 10 to 20C lower with v26.6 than with v28.5.

Please repeat the Blend and Large FFT's tests. It's very likely that you won't have any "Rounding" errors. If you do, then we'll increase memory voltage by .025.

Please read this Tom’s Sticky: Intel Temperature Guide - http://www.tomshardware.com/forum/id-1800828/intel-temperature-guide.html

CT :sol:
 
Upvote 1 Downvote
Solution
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530


Hmm I did not know that about Prime95. OK well I will start those tests respectively and get back with the results.

I must say yes I was getting high temperatures, in the neighborhood of 66C peaking. But I must say isn't that the point? Pushing these devices to their limit and see if they give? That why I run these 24 torcher tests, I want totally stable no matter how unrealistic the circumstances.
 
Upvote 0 Downvote
Zircoben

Zircoben

Distinguished
Apr 15, 2014
1,474
0
20,710


Holy cow, I did NOT know that, that's great to know!

Why did Intel move away from soldered IHS's? Cost?
 
Upvote 0 Downvote
CompuTronix

CompuTronix

Intel Master
Moderator
Nov 13, 2006
6,374
279
38,240
Just FYI, the description for Large FFT's is somewhat misleading.

Where it says "maximum heat, power consumption" actually refers to total heat and power among the processor and memory, not just the processor individually.

Your highest Core temperatures will occur during Small FFT's.
 
Upvote 0 Downvote
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530


Thats good to know, thanks. But sadly during the blend test it did infact have a fatal error in worker 2. "Rounding was 0.5, expected less then 0.4" after 16h 29m. The CPU temp is 52C and the peak was 64C.

I have to ask do you think this is a thermal issue. I am using the Intel stock cooler and I have low opinion of that cooler. If an aftermarket cooler would bring down the temps and stop the errors, I can buy one nbd.
 
Upvote 0 Downvote
CompuTronix

CompuTronix

Intel Master
Moderator
Nov 13, 2006
6,374
279
38,240
How long did it take for Blend to error?

It's not typically a thermal issue until Core temperatures reach 80C Since you're running a non-K processor, the stock cooler is adequate unless you run apps which use AVX / FMA for encoding, such as Handbrake, which can drive Core temperatures into the 90's at stock settings.

I understand that stability testing is a subjective and controversial topic, however, many enthusiasts run P95 tests anywhere from 2 to 8 hours, or perhaps even 12 hours, but 24 hours is extreme. To put it into perspective, if you don't run heavy crunching tasks overnight, then your rig needs only to test stable for as long as you can sit and run a continuous Windows session.

This is of course just my opinion as an electronics engineer, an experienced troubleshooter, computer builder and overclocker. Personally, I've never run any P95 tests past 12 hours.

Nonetheless, it's possible that one of your memory modules has an intermittent component which requires many hours under stress to GLITCH (Good Lord I Thought that Couldn't Happen). :D Intermittent is a 12 letter 4 letter word! :D

You may have to resort to running Blend on one module at a time ... a long and tedious process ... but a solid troubleshooting technique.

In your first post it appears that you mistook Processor Core voltage for memory voltage, did you ever go back into BIOS and raise the memory voltage?

CT :sol:
 
Upvote 0 Downvote
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530


I know there are alot of opinions about stress testing, its very subjective, I know. I I have talked to peole such as your self who say I'm on the high end, other people say I'm on the low end. XD I once talked to a guy who wasn't convinced of his computers stability until he ran Prime95 for 8 months straight.

I would never let my computer run at 80C, I honestly don't like to see temps too far above 40C under load. I typically water cool or put pretty massive heat sinks on all of my CPUs, along with high dollar thermal paste and all that good stuff. I am not putting a big cooler on this computer because I know what the person who ordered this does with their computer. Its simply not necessary, so I didn't plan for it but its an option. I am uncomfortable with the temps I'm currently seeing at 60C but I know thats not enough to actually break anything.

I went into the BIOS to up the RAM voltage from 1.5V to 1.525V and noticed that the speed was autodetecting incorrectly. So I corrected the speed (its DDR3-1600 it detected DDR3-1333) and upped the voltage to the RAM up 0.025V as per your recommendation.

Running the blend test again, see how it plays out. :) Thanks for all of your help so far by the way.
 
Upvote 0 Downvote
CompuTronix

CompuTronix

Intel Master
Moderator
Nov 13, 2006
6,374
279
38,240
You're welcome.

Just another FYI, Intel's "Tcase" Thermal Specification is CPU temperature, not Core Temperature. Core temperature is 5C higher than CPU temperature due to sensor location.

The relationship between Core temperature and CPU temperature is not in the Thermal Specifications; it's only found in a few engineering documents. Tcase for the i5 4590 is 72C. Tcase + 5 makes the corresponding Core temperature 77C.

Core temperatures above show that mid 70's are safe.

CT :sol:
 
Upvote 0 Downvote
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530


Alright after a bunch more testing all things seem stable. You're awesome for sticking with me, thanks so much! :D
 
Upvote 0 Downvote
umdoobby

umdoobby

Honorable
Jul 26, 2012
35
0
10,530




Yea, seem to run perfectly through all of my tests. Thanks for your help!
 
Upvote 0 Downvote
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom