overclocking an i7-2600 to finish an 100M exponent in less than a year :)

emily · 2013-02-28, 18:40

I decided to LL the 100M M332277559 exponent on my quad-core i7-2600. It runs at 3.5GHz without overclock when all cores are 100%. I found out that the best work assignment was to use 3 cores for M332277559 and then use the last fourth core for LL on a normal exponent, M57913199. Assigning 4 cores for the big exponent didn't help much, and letting a core idle wasn't acceptable :)

But at 3.5GHz the M332277559 LL test would take more than a year, so I decided to overclock. My i7 isn't a K so I cannot go very high. I got to 3.9GHz using MSI's OCGenie. But I wanted more, so I went to 4GHz and lastly to 4.1GHz by increasing the BCLK to 105.2 MHz (from 100 MHz). Running at 105.7 MHz is totally unstable, but at 105.5 MHz it looks stable, I decided to let it run at 105.2 MHz to make sure it's stable enough.

At 4.1 GHz with the stage 2 testing 80% complete (I assigned 4GB RAM to it, out of my 16GB total), M332277559 will be completed in November :-)

I also tried to make the 1600MHz DDR3 RAM go faster by adjusting its timings to 8-8-8 rather than 9-9-9 (the 1600MHz RAM modules I use are rated at 8-8-8) but the machine was totally unstable so I went back to 9-9-9.

Power use of the PC at 100% is about 230W (I've got a watt meter hehe! this includes 16W for the 24" 1080p monitor).

It's my first 100M exponent so I'm excited :) but I want to make it go faster! I've got a budget of 200euro which is roughly the price of an i5-3570K which is overclockable and compatible with my motherboard, but it has only 6MB cache instead of 8MB (yet, I don't use HyperThreading, I've it disabled). My current temps are about 52C with the CoolerMaster 612S cooler and two 120mm heatsink fans. But if I get the i5 now then I'll have nothing to do with the i7 chip as I don't have a second motherboard/PC for it, I'll have to wait until next month to turn the i7 into a full PC :>

So here the question: get the i5 now (to speed-up the 100M exponent) and let i7 wait a month for its motherboard/case/etc? or wait until I can get the i5 and the extra components for the i7 at the same time? :)

and: I know prime doesn't care much about 6MB L3 cache vs 8MB, but what about other programs? like Photoshop etc? Will i5 with its 6MB cache be slower than i7 with 8MB for non-prime work?

TheJudger · 2013-02-28, 18:57

Quote:

Originally Posted by emily

Running at 105.7 MHz is totally unstable, but at 105.5 MHz it looks stable, I decided to let it run at 105.2 MHz to make sure it's stable enough.

I would choose a much higher safety margin.

henryzz · 2013-02-28, 18:59

Overclocking is fair enough for a relatively short test(max of a few weeks) but for a year long test the chance of an error is very high.

retina · 2013-02-28, 19:07

Well I guess I'm going to bring you down a little and suggest that your chance of ever completing a 100Mdigit exponent without an error is quite slim (especially when you are overclocking) unless you have ECC RAM installed and enabled.

Without ECC: most probably you will get an error, or ten, over the course of the computation. Sometimes the errors are detectable and sometimes they are not. You would be taking a large chance at hoping all the errors will be both detectable and recomputed successfully each and every time.

Computing 100Mdigit Mersenne residues is no easy task, and without the right hardware, very unlikely to complete successfully.

PhilF · 2013-02-28, 20:11

I agree. There is no way the result will be good.

I did a lot of overclocking, and here is what I found:

Find the edge, just like you did, and back off a little. It "seems" stable. But if you let prime 95's stress test run long enough, you will indeed encounter an error.

At that point, the more you back off the speed, the longer you get to go before the next error. Days, then weeks, then even months. With the test you are running, even one error in 6 months is not acceptable.

In my opinion you are going to have to go down to around 3.7 to 3.8 Ghz before you even stand a chance (unless you are willing to turn up voltages, increase cooling, spend more for electricity, etc). At that point you have to even consider whether or not it is even worth overclocking.

2013-02-28, 18:40	#1
emily Feb 2012 Athens, Greece 47₁₀ Posts	overclocking an i7-2600 to finish an 100M exponent in less than a year :) I decided to LL the 100M M332277559 exponent on my quad-core i7-2600. It runs at 3.5GHz without overclock when all cores are 100%. I found out that the best work assignment was to use 3 cores for M332277559 and then use the last fourth core for LL on a normal exponent, M57913199. Assigning 4 cores for the big exponent didn't help much, and letting a core idle wasn't acceptable :) But at 3.5GHz the M332277559 LL test would take more than a year, so I decided to overclock. My i7 isn't a K so I cannot go very high. I got to 3.9GHz using MSI's OCGenie. But I wanted more, so I went to 4GHz and lastly to 4.1GHz by increasing the BCLK to 105.2 MHz (from 100 MHz). Running at 105.7 MHz is totally unstable, but at 105.5 MHz it looks stable, I decided to let it run at 105.2 MHz to make sure it's stable enough. At 4.1 GHz with the stage 2 testing 80% complete (I assigned 4GB RAM to it, out of my 16GB total), M332277559 will be completed in November :-) I also tried to make the 1600MHz DDR3 RAM go faster by adjusting its timings to 8-8-8 rather than 9-9-9 (the 1600MHz RAM modules I use are rated at 8-8-8) but the machine was totally unstable so I went back to 9-9-9. Power use of the PC at 100% is about 230W (I've got a watt meter hehe! this includes 16W for the 24" 1080p monitor). It's my first 100M exponent so I'm excited :) but I want to make it go faster! I've got a budget of 200euro which is roughly the price of an i5-3570K which is overclockable and compatible with my motherboard, but it has only 6MB cache instead of 8MB (yet, I don't use HyperThreading, I've it disabled). My current temps are about 52C with the CoolerMaster 612S cooler and two 120mm heatsink fans. But if I get the i5 now then I'll have nothing to do with the i7 chip as I don't have a second motherboard/PC for it, I'll have to wait until next month to turn the i7 into a full PC :> So here the question: get the i5 now (to speed-up the 100M exponent) and let i7 wait a month for its motherboard/case/etc? or wait until I can get the i5 and the extra components for the i7 at the same time? :) and: I know prime doesn't care much about 6MB L3 cache vs 8MB, but what about other programs? like Photoshop etc? Will i5 with its 6MB cache be slower than i7 with 8MB for non-prime work? Last fiddled with by emily on 2013-02-28 at 18:44

2013-02-28, 19:07	#4
retina Undefined "The unspeakable one" Jun 2006 My evil lair 2²·1,549 Posts	Well I guess I'm going to bring you down a little and suggest that your chance of ever completing a 100Mdigit exponent without an error is quite slim (especially when you are overclocking) unless you have ECC RAM installed and enabled. Without ECC: most probably you will get an error, or ten, over the course of the computation. Sometimes the errors are detectable and sometimes they are not. You would be taking a large chance at hoping all the errors will be both detectable and recomputed successfully each and every time. Computing 100Mdigit Mersenne residues is no easy task, and without the right hardware, very unlikely to complete successfully. Last fiddled with by retina on 2013-02-28 at 19:09

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
100M exponent FFT size?	NookieN	Information & Answers	3	2017-09-04 22:11
how to submit manually a job with exponent 100M which is done by mfaktc?	fairsky	Information & Answers	17	2013-09-16 19:49
Will Any Current 100M Digit LL Tests Finish?	jinydu	Lounge	121	2012-05-03 15:15
100M exponent idea	MrOzzy	Information & Answers	40	2008-11-12 10:19
Need a way to finish exponent and have it automatically quit.	jasong	Software	4	2007-02-21 23:13

2013-02-28, 18:59	#3
henryzz Just call me Henry "David" Sep 2007 Cambridge (GMT/BST) 2³·3·5·7² Posts	Overclocking is fair enough for a relatively short test(max of a few weeks) but for a year long test the chance of an error is very high.

2013-02-28, 20:11	#5
PhilF Feb 2005 Colorado 2²·7·23 Posts	I agree. There is no way the result will be good. I did a lot of overclocking, and here is what I found: Find the edge, just like you did, and back off a little. It "seems" stable. But if you let prime 95's stress test run long enough, you will indeed encounter an error. At that point, the more you back off the speed, the longer you get to go before the next error. Days, then weeks, then even months. With the test you are running, even one error in 6 months is not acceptable. In my opinion you are going to have to go down to around 3.7 to 3.8 Ghz before you even stand a chance (unless you are willing to turn up voltages, increase cooling, spend more for electricity, etc). At that point you have to even consider whether or not it is even worth overclocking.