Identical PenD Dual-Core CPUs, one is 2.25x faster

NBtarheel_33 · 2008-11-04, 12:39

I recently was able to borg a couple of dual-core Pentium D machines (3.40 ghz). Both machines are working on exponents of comparable size (one for each core, size 29-30M range), but one machine is putting out about 1.8 million LL iterations (x2) each day, while the other machine is putting out only 800-900k LL iterations (x2) each day. At this rate, the slow machine will take over 30 days to test a 29,9xx,xxx exponent, even though PrimeNet's estimated ETA is more like 20 days (which is more like what the first machine will accomplish). This disparity also shows up in trial factoring - 66->67 takes almost 10 hours on the slowpoke, and 4-5 on the "normal" machine.

These CPUs, as far as I can tell, are identical. Ditto for L2 and RAM. As far as I can see, there are no applications that seem to be stealing cycles. No screen saver. There may be some slight use during the daytime hours, but nothing that should cut performance by more than half. Both systems are on 24x7.

What could be happening here? Is this a sign that the slow system is on its way out? Could there be something running that I can't see, that is eating CPU cycles? I haven't seen any error codes in results.txt, so I wouldn't think there'd be any major problems. Just very strange to have two identical systems performing so differently.

Thoughts, opinions, &c. are appreciated.

db597 · 2008-11-04, 15:06

The P4Ds run very hot - my theory would be that the slower one might be throttling the clockspeed due to overheating. It's worth checking the temperatures - you may need to clean and re-seat the heatsink.

petrw1 · 2008-11-04, 15:13

Quote:

Originally Posted by db597

The P4Ds run very hot - my theory would be that the slower one might be throttling the clockspeed due to overheating. It's worth checking the temperatures - you may need to clean and re-seat the heatsink.

Or simply blow the dust off the fans and vents

NBtarheel_33 · 2008-11-04, 19:44

Bet that's it, since the computers are at the front of classrooms with good old fashioned chalkboards that are heavily used (it's the math department, after all). There is a healthy coating of chalk dust all over everything, so I bet the fans and vents are pretty jammed up as well. (You'd think they'd switch to whiteboards or something to avoid having their computers fry themselves!)

So, I will see what I can do to clear the dust away from all openings, and see if that makes any difference. Thanks for your suggestions!

NBtarheel_33 · 2008-11-08, 09:23

I'm happy to report that this had a simple fix. Somehow, the Worker Windows setting was set up to have both exponents running on CPU #1! So what was basically happening was that performance was being reduced to that of one hyperthreaded P4 CPU. As soon as I fixed the CPU affinities, both exponents started running at normal speed.

db597 · 2008-11-12, 03:24

Ah... that's great news, glad you got the issue pinned down. I normally leave mine on the default setting - smart assignment.

I just had one of my PentiumD systems die on me last month (one of the capacitors on the motherboard was bloated). It was overclocked from 3GHz to 3.2GHz and had been running Prime95 24/7 for almost 3 years. Was a great space heater for those cold winter nights.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
How to retire one core in a dual-core CPU?	Rodrigo	PrimeNet	4	2011-07-30 14:43
Dual Core to Quad Core Upgrade	Rodrigo	Hardware	6	2010-11-29 18:48
Importance of dual channel memory for dual core processors	patrik	Hardware	3	2007-01-07 09:26
Dual Core? BFD	R.D. Silverman	Hardware	12	2005-02-20 21:46
Dual CPUs and Hyperthreading	Unregistered	Hardware	34	2004-09-27 08:56

2008-11-04, 12:39	#1
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5·223 Posts	Identical PenD Dual-Core CPUs, one is 2.25x faster I recently was able to borg a couple of dual-core Pentium D machines (3.40 ghz). Both machines are working on exponents of comparable size (one for each core, size 29-30M range), but one machine is putting out about 1.8 million LL iterations (x2) each day, while the other machine is putting out only 800-900k LL iterations (x2) each day. At this rate, the slow machine will take over 30 days to test a 29,9xx,xxx exponent, even though PrimeNet's estimated ETA is more like 20 days (which is more like what the first machine will accomplish). This disparity also shows up in trial factoring - 66->67 takes almost 10 hours on the slowpoke, and 4-5 on the "normal" machine. These CPUs, as far as I can tell, are identical. Ditto for L2 and RAM. As far as I can see, there are no applications that seem to be stealing cycles. No screen saver. There may be some slight use during the daytime hours, but nothing that should cut performance by more than half. Both systems are on 24x7. What could be happening here? Is this a sign that the slow system is on its way out? Could there be something running that I can't see, that is eating CPU cycles? I haven't seen any error codes in results.txt, so I wouldn't think there'd be any major problems. Just very strange to have two identical systems performing so differently. Thoughts, opinions, &c. are appreciated. Last fiddled with by NBtarheel_33 on 2008-11-04 at 12:43 Reason: add info about trial factoring

2008-11-04, 15:06	#2
db597 Jan 2003 11001011₂ Posts	The P4Ds run very hot - my theory would be that the slower one might be throttling the clockspeed due to overheating. It's worth checking the temperatures - you may need to clean and re-seat the heatsink.

2008-11-04, 19:44	#4
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5×223 Posts	Bet that's it, since the computers are at the front of classrooms with good old fashioned chalkboards that are heavily used (it's the math department, after all). There is a healthy coating of chalk dust all over everything, so I bet the fans and vents are pretty jammed up as well. (You'd think they'd switch to whiteboards or something to avoid having their computers fry themselves!) So, I will see what I can do to clear the dust away from all openings, and see if that makes any difference. Thanks for your suggestions!

2008-11-08, 09:23	#5
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5×223 Posts	I'm happy to report that this had a simple fix. Somehow, the Worker Windows setting was set up to have both exponents running on CPU #1! So what was basically happening was that performance was being reduced to that of one hyperthreaded P4 CPU. As soon as I fixed the CPU affinities, both exponents started running at normal speed.

2008-11-12, 03:24	#6
db597 Jan 2003 7×29 Posts	Ah... that's great news, glad you got the issue pinned down. I normally leave mine on the default setting - smart assignment. I just had one of my PentiumD systems die on me last month (one of the capacitors on the motherboard was bloated). It was overclocked from 3GHz to 3.2GHz and had been running Prime95 24/7 for almost 3 years. Was a great space heater for those cold winter nights.