BOINC?

[Rytis]

As you (may) know, PrimeGrid shares the server with BOINCstats. The problem with the server load generally occurs after 16:30GMT when BOINCstats finishes updating stats and a huge amount of users flows in to check how well they have done during the last day :)

The collocation of the server costs 46.50€, of which PrimeGrid pays 21.50€. This amount is just about covered by Google Ads. The small donations from the users helps to keep a small surplus for unexpected events. Currently there is about 35€ remaining, so that's not really enough for the new server ;)

I think that the current server would be quite sufficient for the project if running alone (without BOINCstats), but the separation means that we should buy the server for BOINCstats and pay the increased collocation fee, which is, in my opinion, a bit too much to handle at this time.

The best would be to win the prize for 10M digits prime (I know, I know, we're not searching for that long numbers... yet ;)) - should solve all financial problems :)

[KEP]

Originally Posted by Rytis

The best would be to win the prize for 10M digits prime (I know, I know, we're not searching for that long numbers... yet ;)) - should solve all financial problems :)

It would be the best to win the prize for the first 10M digit prime, and I think that we can do it. If you wan't to, I could already for now start sieving on an abandoned machine, I guess I'll would sieve it up to 25 G, then maybe those pacionet or whoever helped you split it into 1M ranges, can help you split the file, I think sieving should be done, when done doing TPS on the current scale :)

The new task if decided to look for 10 M digit TPs is going to be: k*2^33,219,278±-1. The n value is based on the mersenne benchmark page, maybe someone can confirm or unconfirm this, since 10000000/log(2) gives n to be approximately 3 higher. Anyone?

Conclusively I'll say that if anyone can do it, PrimeGrid can do it, and also I expect to be able to do 1 task every 8 day :)

[victor]

KEP, how long does LLR take to test a 10M digit number?

[MooMoo2]

I don't think we should be searching for a 10 million digit twin prime after we find a prime for n=195000. Sure, it's nice to dream big, but the amount of work in finding a 10 million digit twin is over 800 MILLION times the amount of work needed to find a twin of the form k*2^195000+/-1. The chance of finding even a single 10 million digit prime (non-twin) isn't great either, because it takes twice as much time to test k*2^(33.2M)-1 than to test 2^(33.2M)-1, due to the bigger k's and larger FFT lengths.

My suggestion would be to progress this way:

- Look for a 100,000 digit twin
- Look for a twin with an exponent >500,000
- Look for a 200,000 digit twin
- Look for a twin with a megabit exponent (k*2^1,000,000+/- 1)
- Look for a 500,000 digit twin
- Look for a twin with exponent >2,000,000
- Look for a million digit twin....
...and so on

It may seem like slow progress, but each step takes an average of 4 times as much computing power as the previous step.

[KEP]

Originally Posted by victor

KEP, how long does LLR take to test a 10M digit number?

I haven't done any testing on this, but mersenne.org, suggest that it would take my Pentium 4 2553 Mhz, approximately 26 days 12 hours and 18 minutes to do a, so I guess that one test should be possible to do in no more than 8 days and 12 hours crunching or in one third the time :)

To give you an example, the given n (195,000) is taking with prime95 a suggest 1 hour 15 minutes, while LLR does 1 test in 1 minutes 55 seconds on my machine or about 39 times faster than suggested. Maybe a test on my computer will take 4-8 days, I actually don't think it is possible for my computer to do it faster, but if anyone is up for it, then please let us all know what time it took for you to test 1 10M digit prime candidate.

So to answer your question, the answer is "Less than 8 days (196 hours)".

[victor]

Originally Posted by KEP

So to answer your question, the answer is "Less than 8 days (196 hours)".

No, it's not. You can't compare these results.
We know that :
-the lucas-lehmer test for mersenne numbers is the fastest known primality test
-prime95 is the fastest known LLT implementation
-the lucas-lehmer test for mersenne numbers doesn't work for numbers of the form k*2^n-1 if k != 1, and doesn't work for numbers of the form k*2^n+1

Now, explain me how do you expect to test a number of the form k*2^n+/-1 in 196 hours without using a supercomputer?

[KEP]

Originally Posted by victor

No, it's not. You can't compare these results.
We know that :
-the lucas-lehmer test for mersenne numbers is the fastest known primality test
-prime95 is the fastest known LLT implementation
-the lucas-lehmer test for mersenne numbers doesn't work for numbers of the form k*2^n-1 if k != 1, and doesn't work for numbers of the form k*2^n+1

Now, explain me how do you expect to test a number of the form k*2^n+/-1 in 196 hours without using a supercomputer?

I just started a for fun test, to see exactly how long it would take, and apparently it will take 99 milliseconds to do 1 iteration, which means that it on my machine will take around 38 days ± ½ day, so as it turns out, LLR is faster on 10 M digit candidates than MPrime is, even though the exact is opposite with the n we are currently testing :) Yeehhh my bad :) Apparently the range though is pretty fast to sieve, so plenty of work should be around for long time :) I might give it a try see how it goes with the 25 G range, if I try to sieve it to 100 Billion :)

[pacionet]

Originally Posted by MooooMoo

I don't think we should be searching for a 10 million digit twin prime after we find a prime for n=195000. Sure, it's nice to dream big, but the amount of work in finding a 10 million digit twin is over 800 MILLION times the amount of work needed to find a twin of the form k*2^195000+/-1. The chance of finding even a single 10 million digit prime (non-twin) isn't great either, because it takes twice as much time to test k*2^(33.2M)-1 than to test 2^(33.2M)-1, due to the bigger k's and larger FFT lengths.

My suggestion would be to progress this way:

- Look for a 100,000 digit twin
- Look for a twin with an exponent >500,000
- Look for a 200,000 digit twin
- Look for a twin with a megabit exponent (k*2^1,000,000+/- 1)
- Look for a 500,000 digit twin
- Look for a twin with exponent >2,000,000
- Look for a million digit twin....
...and so on

It may seem like slow progress, but each step takes an average of 4 times as much computing power as the previous step.

I agree. Let's start from the simpler things.

[Rytis]

Hey, I just stated it would be cool to win :) I however understand that it's not possible with current level of technology.

On a side note, with a bigger influx of users we're sending out work fast. New users already found 2 primes. I'll probably be forced to insert more work tomorrow, and I created a script to automate the process without the need to split the files into 1M parts. I reserved a range of 90M (1010M-1100M), and it would be great if someone (gribozavr?) could provide me with a single sieved file and not 18 as I would need now :) Thanks.

[victor]

victor@kortchnoi ~/Desktop/tps $ ls
1010e6-1015e6_195000.txt  1040e6-1045e6_195000.txt  1070e6-1075e6_195000.txt
1015e6-1020e6_195000.txt  1045e6-1050e6_195000.txt  1075e6-1080e6_195000.txt
1020e6-1025e6_195000.txt  1050e6-1055e6_195000.txt  1080e6-1085e6_195000.txt
1025e6-1030e6_195000.txt  1055e6-1060e6_195000.txt  1085e6-1090e6_195000.txt
1030e6-1035e6_195000.txt  1060e6-1065e6_195000.txt  1090e6-1095e6_195000.txt
1035e6-1040e6_195000.txt  1065e6-1070e6_195000.txt  1095e6-1100e6_195000.txt
victor@kortchnoi ~/Desktop/tps $ grep 1: 1010e6-1015e6_195000.txt >> 1010e6-1100 e6_195000.txt && cat * | grep -v 1: >> 1010e6-1100e6_195000.txt
victor@kortchnoi ~/Desktop/tps $

And voila :)

[pacionet]

Wow !
Gribozavr we need next pre-sieved ranges 1100-1200M !!!!

Thanks
Andrea

You (or Victor) can send me the "All in one" file with all the 100M range (for Primegrid) ?
I 'll post them here, in "All in one", bottom of the page:

http://www.twinprimesearch.org/pre-sieved_ranges.htm

Thanks
Andrea