Haswell Preview Benchmark - Page 26

db597 · 2013-09-10, 13:59

Many motherboards have the option to disable it. On my Asus Z87 motherboard, if you select adaptive voltage, it will bump by 0.1V when AVX/FMA3 are used.

As you say, 0.1V is quite a lot and it makes overclockers who are already using elevated voltages feel uncomfortable. In such a case, you can select manual voltage and this will keep a constant voltage.

I had a BSOD again after almost 24hrs uptime. Now bumped voltage to 1.275V @ 4.2GHz for the core, and for the cache running at 1.26V @ 4.0GHz.

db597 · 2013-09-10, 14:08

Quote:

Originally Posted by db597

I had a BSOD again after almost 24hrs uptime. Now bumped voltage to 1.275V @ 4.2GHz for the core, and for the cache running at 1.26V @ 4.0GHz.

Update: Hitting 80 degrees, even with a Corsair H110 watercooler... grr.. maybe it's more sane to try a lower overclock afterall.

TheMawn · 2013-09-11, 04:46

Quote:

Originally Posted by db597

Update: Hitting 80 degrees, even with a Corsair H110 watercooler... grr.. maybe it's more sane to try a lower overclock afterall.

If you enjoy that sort of thing, I might try re-applying the cooler in your shoes. My i5-3570k is at 4.6 GHz at 1.28 volts, and runs roughly 50C above ambient. I use an H80i with two Noctua 120mm fans which actually operate silently.

Are you using two or four 140mm fans? Either way you should be getting a LOT more cooling power than me. Lower frequency and same voltage, and still getting ~10C above me?

What memory do you use?

henryzz · 2013-09-11, 14:21

Quote:

Originally Posted by db597

I had a BSOD again after almost 24hrs uptime. Now bumped voltage to 1.275V @ 4.2GHz for the core, and for the cache running at 1.26V @ 4.0GHz.

Do you get any performance boost from running the core and clock at the same speed? I would imagine that if the cpu recognized that and locked them together then it could provide some improvement. Similar things like running memory at 2x fsb have been reckoned to provide small boosts in the past.

tomaz · 2013-09-11, 15:00

I am getting an error in prime95 v28.1. It always pops up after 1 hour and 7-9 minutes of torture small FFTs.
The error first appears on one worker and then on all the other workers.
I am running it on 4670K at 4300MHz. Have tested lots of different settings from 1.24 to 1.28 VCore, different Vrin, Vring, ...; always all workers stop at the same test.

Is it possible that this is software bug, since I am "failing the torture test in the SAME SPOT with the SAME ERROR MESSAGE"?
Is there a simple way to set all workers to start a this test:
"Test 2, 1500000 Lucas-Lehmer iterations of M262143 using FMA3 FFT length 15K, Pass1=320, Pass2=48"
?

[Sep 10 20:23] Self-test 8K passed!
[Sep 10 20:23] Test 1, 1500000 Lucas-Lehmer iterations of M266241 using FMA3 FFT length 15K, Pass1=320, Pass2=48.
[Sep 10 20:24] Test 2, 1500000 Lucas-Lehmer iterations of M262143 using FMA3 FFT length 15K, Pass1=320, Pass2=48.
[Sep 10 20:25] FATAL ERROR: Final result was 7A473DE6, expected: 00000000.
[Sep 10 20:25] Hardware failure detected, consult stress.txt file.
[Sep 10 20:25] Torture Test completed 63 tests in 1 hour, 8 minutes - 1 errors, 0 warnings.
[Sep 10 20:25] Worker stopped.

TheMawn · 2013-09-11, 17:32

Quote:

Originally Posted by db597

I had a BSOD again after almost 24hrs uptime. Now bumped voltage to 1.275V @ 4.2GHz for the core, and for the cache running at 1.26V @ 4.0GHz.

I just caught this. Wasn't really paying attention. Just saw 1.275 V 4.2 GHz 1.26 V 4.0 GHz.

I have never heard of a cache speed...

Prime95 · 2013-09-11, 17:35

Quote:

Originally Posted by tomaz

Is it possible that this is software bug, since I am "failing the torture test in the SAME SPOT with the SAME ERROR MESSAGE"?

Yes. Version 28 has all new self-test data (more iterations so that less time is spent in setup and more time doing iterations). It looks like I failed to properly set the expected residue for that exponent. The big clue is the "expected 00000000" string.

Thanks for the bug report.

TheMawn · 2013-09-11, 17:41

If your result is wrong, then you have just gotten an error somewhere in the test. The work is iterative: You run the algorithm on the number four, then on the result of that algorithm, then on the result of that algorithm, etc. Your hardware might be making a mistake just a few hundred iterations into the test, which apparently has one and a half million iterations.

The result that was expected being a bunch of zeros indicates the exponent should be prime. Your result indicates that it is not. Check that the result you actually get is the same every single time. Not just the form of the message, but the actual numbers. The 7A473DE6 bit.

If that part is the same every time, you have some sort of systematic instability...

Try a blend test or something as well. It'll give you a bigger variety of stuff to try.

TheMawn · 2013-09-11, 17:42

Quote:

Originally Posted by Prime95

Yes. Version 28 has all new self-test data (more iterations so that less time is spent in setup and more time doing iterations). It looks like I failed to properly set the expected residue for that exponent. The big clue is the "expected 00000000" string.

Thanks for the bug report.

... or there's actually a bug.

tomaz · 2013-09-11, 18:51

Quote:

Originally Posted by Prime95

Yes. Version 28 has all new self-test data (more iterations so that less time is spent in setup and more time doing iterations). It looks like I failed to properly set the expected residue for that exponent. The big clue is the "expected 00000000" string.

Thanks for the bug report.

You're welcome. Thanks for the v28.1.

Prime95 · 2013-09-11, 19:09

Here is the Windows 64-bit version that corrects the bad self-test data:

http://www.sendspace.com/file/mtfcor

2013-09-11, 19:09	#286
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 19·397 Posts	Prime95 version 28.1 build 2 Here is the Windows 64-bit version that corrects the bad self-test data: http://www.sendspace.com/file/mtfcor

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Haswell-E Prelim. Benchmark	sdbardwick	Hardware	37	2015-02-10 18:49
Prime95 and Haswell	Pleco	Information & Answers	22	2014-07-13 16:03
Haswell Rig	Mini-Geek	Hardware	64	2014-05-27 13:22
Prime95 version 27.1 early preview, not-even-close-to-beta release	Prime95	Software	126	2012-02-09 16:17
Missing mouse-over preview text	retina	Forum Feedback	1	2011-09-12 15:32

2013-09-10, 13:59	#276
db597 Jan 2003 7×29 Posts	Many motherboards have the option to disable it. On my Asus Z87 motherboard, if you select adaptive voltage, it will bump by 0.1V when AVX/FMA3 are used. As you say, 0.1V is quite a lot and it makes overclockers who are already using elevated voltages feel uncomfortable. In such a case, you can select manual voltage and this will keep a constant voltage. I had a BSOD again after almost 24hrs uptime. Now bumped voltage to 1.275V @ 4.2GHz for the core, and for the cache running at 1.26V @ 4.0GHz.

2013-09-11, 15:00	#280
tomaz Sep 2013 3 Posts	I am getting an error in prime95 v28.1. It always pops up after 1 hour and 7-9 minutes of torture small FFTs. The error first appears on one worker and then on all the other workers. I am running it on 4670K at 4300MHz. Have tested lots of different settings from 1.24 to 1.28 VCore, different Vrin, Vring, ...; always all workers stop at the same test. Is it possible that this is software bug, since I am "failing the torture test in the SAME SPOT with the SAME ERROR MESSAGE"? Is there a simple way to set all workers to start a this test: "Test 2, 1500000 Lucas-Lehmer iterations of M262143 using FMA3 FFT length 15K, Pass1=320, Pass2=48" ? [Sep 10 20:23] Self-test 8K passed! [Sep 10 20:23] Test 1, 1500000 Lucas-Lehmer iterations of M266241 using FMA3 FFT length 15K, Pass1=320, Pass2=48. [Sep 10 20:24] Test 2, 1500000 Lucas-Lehmer iterations of M262143 using FMA3 FFT length 15K, Pass1=320, Pass2=48. [Sep 10 20:25] FATAL ERROR: Final result was 7A473DE6, expected: 00000000. [Sep 10 20:25] Hardware failure detected, consult stress.txt file. [Sep 10 20:25] Torture Test completed 63 tests in 1 hour, 8 minutes - 1 errors, 0 warnings. [Sep 10 20:25] Worker stopped.

2013-09-11, 17:41	#283
TheMawn May 2013 East. Always East. 11·157 Posts	If your result is wrong, then you have just gotten an error somewhere in the test. The work is iterative: You run the algorithm on the number four, then on the result of that algorithm, then on the result of that algorithm, etc. Your hardware might be making a mistake just a few hundred iterations into the test, which apparently has one and a half million iterations. The result that was expected being a bunch of zeros indicates the exponent should be prime. Your result indicates that it is not. Check that the result you actually get is the same every single time. Not just the form of the message, but the actual numbers. The 7A473DE6 bit. If that part is the same every time, you have some sort of systematic instability... Try a blend test or something as well. It'll give you a bigger variety of stuff to try.