Hyperthreading problems...

Xyzzy · 2003-08-26, 22:47

I can run one instance of the torture test forever with no errors... If I run two copies one will fail within minutes...

One torture test running...
http://www.mersenneforum.org/jpg/tt1.jpg

Two torture tests running...
http://www.mersenneforum.org/jpg/tt2.jpg

As you can see, running two hits the CPU much harder... It tends to run a few degrees hotter too...

I'm using 23.6 and I set the test to "blend"...

The error...

Quote:

[Tue Aug 26 17:39:30 2003]
FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.

Xyzzy · 2003-08-26, 22:54

The hardware is not overclocked... It is a 2.4C on an Intel board... The memory is running from the SPD settings...

The only problem I can see so far is the voltages on the power supply are a bit low for the vCore and vIO... I've tested three different power supplies and they all read the same...

vCore = 1.454 (1.525)
vIO = 1.470 - 1.483 (1.500)
3.3v = 3.310 (3.300)
5v = 5.078 - 5.105 (5.000)
12v = 12.000 - 12.062 (12.000)

The number in parenthesis is the optimal number...

The box passes the extended memtest86 tests...

I can also get the error if I run a different DC project and Prime95 at the same time... The other project won't fail, or course, since they usually don't have error checking, but Prime95 sure will... And not only does the torture test fail, so does ECM factoring, with the same sumout error...

My guess is the CPU and not the memory...

Prime Monster · 2003-08-26, 22:58

I believe the Vcore is your problem. Try to up it to 1.500

heretic

Xyzzy · 2003-08-26, 23:01

The vCore is within Intel's guidelines... (There is no option on an Intel motherboard to adjust it anyways!)

It is 4.7% low... IIRC, the spec is within 5%...

CPU info...

http://processorfinder.intel.com/scripts/details.asp?sSpec=SL6WF

Also, look at page 23...

ftp://download.intel.com/design/Pentium4/datashts/29864310.pdf

Xyzzy · 2003-08-26, 23:32

Quote:

Originally Posted by Xyzzy

I'm using 23.6 and I set the test to "blend"...

Oops, I'm using "in-place large FFTs"...

Next I am going to run two torture tests with Hyperthreading turned off...

Xyzzy · 2003-08-27, 00:20

With it turned off, the torture test ran 45 minutes with 2 instances of it running... Even with 1 instance the CPU meter was pegged...

I'm going to temporarily replace the Kingston memory that is installed with 2 new sticks of Crucial DDR400 and then I'll test it with Hyperthreading turned on... Maybe this will eliminate the memory as the culprit...

Prime95 · 2003-08-27, 01:16

Try using memory settings slower than SPD.

Xyzzy · 2003-08-27, 01:52

It just passed an hour with the Crucial memory... I'm testing 2 of the original sticks by themselves now, and if they go an hour then I'll try the other 2... If they all pass that way I'll try all 4 at a slower speed...

Possibly they might be overheating when they are all installed. I mention this because the heat spreaders on the sticks in the same channel touch each other... With just 1 stick per channel there is a gap...

Xyzzy · 2003-08-27, 09:12

Well, all possible combinations of 2 sticks passed... I guess that means either the motherboard can't handle 4 sticks or maybe they are overheating due to the fact they are packed in so tight...

Tomorrow I'll run a test with the memory at DDR333...

Xyzzy · 2003-08-27, 20:02

Boy this is a lot of work... :(

I aimed a fan this morning at the memory and so far it is passing the torture test (2 instances) going on 6 hours...

With a single stick per channel they run 40-42C, but when I put both sticks in per channel they run 50-52C... The south bridge is 45C, the voltage regulators are 60C and the north bridge is 40-45C...

George-

When I run 1 instance of the torture test, or LL testing, on the box with Hyperthreading turned on, the graph (above) shows that the CPU is not being used fully... When I kick on 2 instances it shows it being used fully and the temp goes up a few degrees... Is it possible to code Prime95 to use those extra cycles?

Finally, I did find, even with an Intel-branded board, that the audio driver would BSOD the box when I ran an ECM test on 67108864, which uses a *lot* of memory... This error is not related to the torture test error I think, but just to be safe I will buy a dedicated sound card today and disable the onboard one...

This is surprising to me because I've never had any driver issues with an Intel-branded board before... :(

Xyzzy · 2003-08-28, 01:40

I just went out and got an Asus P4P800-Deluxe motherboard... Just for fun I'll see if I can duplicate this error...

(Does anyone need an Intel D865GLC cheap?)

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Hyperthreading	TheMawn	Hardware	12	2013-08-15 00:03
Hyperthreading	Primeinator	Information & Answers	13	2010-05-20 15:15
Should hyperthreading be used?	Electrolyte	Hardware	5	2006-11-08 01:29
Hyperthreading	dave_0273	Hardware	5	2003-12-12 13:22
Hyperthreading	RickC	Hardware	14	2003-12-04 14:23

2003-08-26, 22:54	#2
Xyzzy Aug 2002 5·11·157 Posts	The hardware is not overclocked... It is a 2.4C on an Intel board... The memory is running from the SPD settings... The only problem I can see so far is the voltages on the power supply are a bit low for the vCore and vIO... I've tested three different power supplies and they all read the same... vCore = 1.454 (1.525) vIO = 1.470 - 1.483 (1.500) 3.3v = 3.310 (3.300) 5v = 5.078 - 5.105 (5.000) 12v = 12.000 - 12.062 (12.000) The number in parenthesis is the optimal number... The box passes the extended memtest86 tests... I can also get the error if I run a different DC project and Prime95 at the same time... The other project won't fail, or course, since they usually don't have error checking, but Prime95 sure will... And not only does the torture test fail, so does ECM factoring, with the same sumout error... My guess is the CPU and not the memory...

2003-08-26, 22:58	#3
Prime Monster Aug 2002 2²×5×13 Posts	I believe the Vcore is your problem. Try to up it to 1.500 heretic

2003-08-26, 23:01	#4
Xyzzy Aug 2002 5·11·157 Posts	The vCore is within Intel's guidelines... (There is no option on an Intel motherboard to adjust it anyways!) It is 4.7% low... IIRC, the spec is within 5%... CPU info... http://processorfinder.intel.com/scripts/details.asp?sSpec=SL6WF Also, look at page 23... ftp://download.intel.com/design/Pentium4/datashts/29864310.pdf

2003-08-27, 00:20	#6
Xyzzy Aug 2002 5·11·157 Posts	With it turned off, the torture test ran 45 minutes with 2 instances of it running... Even with 1 instance the CPU meter was pegged... I'm going to temporarily replace the Kingston memory that is installed with 2 new sticks of Crucial DDR400 and then I'll test it with Hyperthreading turned on... Maybe this will eliminate the memory as the culprit...

2003-08-27, 01:16	#7
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2039₁₆ Posts	Try using memory settings slower than SPD.

2003-08-27, 01:52	#8
Xyzzy Aug 2002 5×11×157 Posts	It just passed an hour with the Crucial memory... I'm testing 2 of the original sticks by themselves now, and if they go an hour then I'll try the other 2... If they all pass that way I'll try all 4 at a slower speed... Possibly they might be overheating when they are all installed. I mention this because the heat spreaders on the sticks in the same channel touch each other... With just 1 stick per channel there is a gap...

2003-08-27, 09:12	#9
Xyzzy Aug 2002 5·11·157 Posts	Well, all possible combinations of 2 sticks passed... I guess that means either the motherboard can't handle 4 sticks or maybe they are overheating due to the fact they are packed in so tight... Tomorrow I'll run a test with the memory at DDR333...

2003-08-27, 20:02	#10
Xyzzy Aug 2002 20673₈ Posts	Boy this is a lot of work... :( I aimed a fan this morning at the memory and so far it is passing the torture test (2 instances) going on 6 hours... With a single stick per channel they run 40-42C, but when I put both sticks in per channel they run 50-52C... The south bridge is 45C, the voltage regulators are 60C and the north bridge is 40-45C... George- When I run 1 instance of the torture test, or LL testing, on the box with Hyperthreading turned on, the graph (above) shows that the CPU is not being used fully... When I kick on 2 instances it shows it being used fully and the temp goes up a few degrees... Is it possible to code Prime95 to use those extra cycles? Finally, I did find, even with an Intel-branded board, that the audio driver would BSOD the box when I ran an ECM test on 67108864, which uses a lot of memory... This error is not related to the torture test error I think, but just to be safe I will buy a dedicated sound card today and disable the onboard one... This is surprising to me because I've never had any driver issues with an Intel-branded board before... :(

2003-08-28, 01:40	#11
Xyzzy Aug 2002 10000110111011₂ Posts	I just went out and got an Asus P4P800-Deluxe motherboard... Just for fun I'll see if I can duplicate this error... (Does anyone need an Intel D865GLC cheap?)