How long will the project take?

lycorn · 2004-07-27, 13:56

I consider the project "finished" when all exponents up to 79.3M have been tested and DCed and/or factored. I know it is difficult to make an educated guess, as the hardware and algorithms will most likely change dramatically in the next couple of years, but I thought it would be interesting to know your feeling as hardcore GIMPSters...

E_tron · 2004-07-27, 15:52

it almost feels like the project is finished right now. we've explored the planets and now we are going into the depts of space.

dave_0273 · 2004-07-27, 16:35

I would probably consider the project "finished" when we find the 10mil digit prime. Unless another cash prize was offered again. I think that there is a cash prize for a 100mil digit prime, but that is too far away (as far as I am concerned) to look for. It would take FAR to long to test a 100mil digit prime with todays hardware (assuming that no new algorithims are developed that are faster than the LL test).

edorajh · 2004-07-27, 18:33

I vote 10-15 years, but everything is possible.

I only hope that GIMPS will not end once we TF, LL and DC all available exponents in current range. I hope we will be able to extend current range and test above current max exponent.

PrimeFun · 2004-07-27, 20:28

I'll be onboard until George kicks the bucket or he losses interest.

Orgasmic Troll · 2004-07-27, 23:56

what exactly is the cause of the limitation? is it something that will be overcome by technology in the next decade or so?

Aside from that .. I'd say we would have to look at how many CPU-years it would take to finish all the exponents in the range and look at the trends of how many CPU-years are completed every year or so. Since I have no idea what either of those are .. I can't give an answer :D

Also, I think it would depend on whether or not GIMPS wins the EFF prize. I think if it does, there will be a marked (but not necessarily dramatic) increase in the number of CPU-years completed every year

Another surprise factor would be whether or not there is a big upheaval in computer technology (along the lines of quantum computing or something)
I'd guesstimate (and I'm sure someone with industry knowledge could give more accurate estimates) that if some new technology is in R+D right now, it'll take 3-5 years to make it commercially viable and 3-5 more years before it makes any impact on GIMPS, and then maybe 2-4 years to make a significant impact, so that's looking at 8-14 years to make a signifcant impact

PrimeCruncher · 2004-07-28, 00:45

Quote:

Originally Posted by TravisT

what exactly is the cause of the limitation? is it something that will be overcome by technology in the next decade or so?

The main barrier is the 79.3M limitation of the Prime95 software, but that will probably be overcome by new algorithms or a replacement of the Lucas-Lehmer test altogether with something faster. Performing an LL test on a billion-digit Mersenne is expected to take 852 years, according to Luigi, on the most modern hardware available. Obviously, with increases in speed and new technologies, this will be reduced.

I find your estimate of R+D to big impact on GIMPS to probably be about right. DDR2 RAM is now finally available, though few people are using it. It'll be a while before lots of people in GIMPS will replace their computers with new DDR2-based machines, so it'll be a few years before that tech has a major impact on GIMPS, as you said. DDR2 should have a large impact; memory performance has historically been a major bottleneck.

aaronl · 2004-07-28, 09:36

Quote:

Originally Posted by PrimeCruncher

The main barrier is the 79.3M limitation of the Prime95 software, but that will probably be overcome by new algorithms or a replacement of the Lucas-Lehmer test altogether with something faster. Performing an LL test on a billion-digit Mersenne is expected to take 852 years, according to Luigi, on the most modern hardware available. Obviously, with increases in speed and new technologies, this will be reduced.

I'm pretty sure the limit is arbitrary. It could be raised in a future release. The limit has nothing to do with the algorithm, only the implementation.

PrimeCruncher · 2004-07-28, 13:27

Quote:

Originally Posted by aaronl

I'm pretty sure the limit is arbitrary. It could be raised in a future release. The limit has nothing to do with the algorithm, only the implementation.

Exactly: the implementation of the alogorithms would have to be changed to support higher exponents. But I don't think the limit is arbitrary. I think it IS a limitation of the way the algorithms are presently implemented. P4s would have issues testing to 79.3M I think, since because they use SSE2 optimizations, they have to switch to higher FFTs earlier than non-SSE2 CPUs.

kd7dvd · 2004-07-28, 17:16

Well, i know we don't yet have a proof that there are an infinite number of Mersenne primes, but ya know, there's lots and lots of candidates. lots and lots and lots. lots and lots and lots and lots...

otoh, i'd be interested in a poll asking how long we think Moore's Law will hold up

Uncwilly · 2013-01-03, 06:45

Reviving this necrothread, rather than starting a new one or tapping into either of these 2 ("Milestones" or "3 years trend").

So, when do you all figure that we will complete LL on all of the exponents under 79.3?

I have been looking at the rates of change in some of the numbers being reported. Looking at the trend of the deltas in P-90 years. And projecting forward, it looks like the P-90 years number could go to zero by the end of March 2015, sometime around July 2015 using a larger data set. Over the long haul the estimated date (with out trend projection) has approached 1/1/2021 several times, but we have been going firmly below that for a while.

View Poll Results: How long will GIMPS take to test all exponents <79.3M?
Less than 10 years	7	17.07%
10 to 15 years	14	34.15%
15 to 20 years	6	14.63%
More than 20 years	14	34.15%
Voters: 41. You may not vote on this poll

2004-07-27, 13:56	#1
lycorn "GIMFS" Sep 2002 Oeiras, Portugal 3056₈ Posts	How long will the project take? I consider the project "finished" when all exponents up to 79.3M have been tested and DCed and/or factored. I know it is difficult to make an educated guess, as the hardware and algorithms will most likely change dramatically in the next couple of years, but I thought it would be interesting to know your feeling as hardcore GIMPSters...

2004-07-27, 18:33	#4
edorajh Oct 2003 Croatia 2³·3·19 Posts	I vote 10-15 years, but everything is possible. I only hope that GIMPS will not end once we TF, LL and DC all available exponents in current range. I hope we will be able to extend current range and test above current max exponent. Last fiddled with by edorajh on 2004-07-27 at 18:41

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
very long int	davar55	Lounge	60	2013-07-30 20:26
Special project #3b - Project 400	schickel	Aliquot Sequences	307	2011-10-28 01:29
Special project #3a - Project 300	schickel	Aliquot Sequences	29	2011-08-12 17:45
Using long long's in Mingw with 32-bit Windows XP	grandpascorpion	Programming	7	2009-10-04 12:13
I think it's gonna be a long, long time	panic	Hardware	9	2009-09-11 05:11

2004-07-27, 15:52	#2
E_tron Sep 2002 Austin, TX 1061₈ Posts	it almost feels like the project is finished right now. we've explored the planets and now we are going into the depts of space.

2004-07-27, 16:35	#3
dave_0273 Oct 2003 Australia, Brisbane 2·5·47 Posts	I would probably consider the project "finished" when we find the 10mil digit prime. Unless another cash prize was offered again. I think that there is a cash prize for a 100mil digit prime, but that is too far away (as far as I am concerned) to look for. It would take FAR to long to test a 100mil digit prime with todays hardware (assuming that no new algorithims are developed that are faster than the LL test).

2004-07-27, 20:28	#5
PrimeFun Jul 2003 2·3·5 Posts	I'll be onboard until George kicks the bucket or he losses interest.

2004-07-27, 23:56	#6
Orgasmic Troll Cranksta Rap Ayatollah Jul 2003 641 Posts	what exactly is the cause of the limitation? is it something that will be overcome by technology in the next decade or so? Aside from that .. I'd say we would have to look at how many CPU-years it would take to finish all the exponents in the range and look at the trends of how many CPU-years are completed every year or so. Since I have no idea what either of those are .. I can't give an answer :D Also, I think it would depend on whether or not GIMPS wins the EFF prize. I think if it does, there will be a marked (but not necessarily dramatic) increase in the number of CPU-years completed every year Another surprise factor would be whether or not there is a big upheaval in computer technology (along the lines of quantum computing or something) I'd guesstimate (and I'm sure someone with industry knowledge could give more accurate estimates) that if some new technology is in R+D right now, it'll take 3-5 years to make it commercially viable and 3-5 more years before it makes any impact on GIMPS, and then maybe 2-4 years to make a significant impact, so that's looking at 8-14 years to make a signifcant impact

2004-07-28, 17:16	#10
kd7dvd May 2004 Seattle, Washington, US 8₁₀ Posts	Well, i know we don't yet have a proof that there are an infinite number of Mersenne primes, but ya know, there's lots and lots of candidates. lots and lots and lots. lots and lots and lots and lots... otoh, i'd be interested in a poll asking how long we think Moore's Law will hold up

2013-01-03, 06:45	#11
Uncwilly 6809 > 6502 """"""""""""""""""" Aug 2003 101×103 Posts 2·3·11·167 Posts	Reviving this necrothread, rather than starting a new one or tapping into either of these 2 ("Milestones" or "3 years trend"). So, when do you all figure that we will complete LL on all of the exponents under 79.3? I have been looking at the rates of change in some of the numbers being reported. Looking at the trend of the deltas in P-90 years. And projecting forward, it looks like the P-90 years number could go to zero by the end of March 2015, sometime around July 2015 using a larger data set. Over the long haul the estimated date (with out trend projection) has approached 1/1/2021 several times, but we have been going firmly below that for a while.