Reservations - Page 5

wombatman · 2017-07-13, 13:45

Ok. I'll let it run for a while and see how things progress. I'll go ahead and reserve n=6201-6299 for k=8e8-9e8 (the first entry in Rogue's list).

wombatman · 2017-07-13, 15:17

Gonna take

Code:

181744-189999 13000 30000

since I can run both of them at the same time with a relatively minimal (<25%) dropoff in prime throughput. Time per factor for this range will need to be ~25 seconds. Time per factor for the previous range needs to be just ~0.1 sec.

rogue · 2017-07-13, 15:48

Thanks wombatman. I'm really curious to see how many candidates remain after sieving.

BTW, what is the format for the output from ppsieve_cuda? I assume it is the pfgw ABCD format. I will likely need to update PRPNet to support that format.

wombatman · 2017-07-13, 15:58

I don't know if it will output a file at the end of a run, but it outputs the factors as

Code:

1000000016401549 | 833778745*2^6246+1

So worst case scenario, you could generate a list, eliminate small factors with srsieve, and then remove the larger factors with srfile.

When the shorter run finishes, I'll let you know whether a file with the remaining candidates is generated.

ET_ · 2017-07-13, 16:10

Quote:

Originally Posted by wombatman

I don't know if it will output a file at the end of a run, but it outputs the factors as

Code:

1000000016401549 | 833778745*2^6246+1

So worst case scenario, you could generate a list, eliminate small factors with srsieve, and then remove the larger factors with srfile.

When the shorter run finishes, I'll let you know whether a file with the remaining candidates is generated.

waitaminnit... you are actually testing for primes at 10¹⁵.
I think you were supposed to start from p=3 (but then, iirc ppsieve needs to start from a p > k), so a whole lot of very small factors are not tested.

When I had a GPU, I started sieving with FermFact, and when p reached the minimum required by ppsieve_CUDA, I saved the results from FermFact and moved them on ppsieve.

How did you start ppsieve without a pre-existing candidates' file?

wombatman · 2017-07-13, 16:37

Quote:

Originally Posted by ET_

waitaminnit... you are actually testing for primes at 10¹⁵.
I think you were supposed to start from p=3 (but then, iirc ppsieve needs to start from a p > k), so a whole lot of very small factors are not tested.

When I had a GPU, I started sieving with FermFact, and when p reached the minimum required by ppsieve_CUDA, I saved the results from FermFact and moved them on ppsieve.

How did you start ppsieve without a pre-existing candidates' file?

You're right--I started with minp=1e15 since ppsieve_cuda wouldn't allow anything smaller than that. I will try running FermFact first and just remove the factors found so far by ppsieve_cuda. Then I will continue onward if appropriate.

ET_ · 2017-07-13, 16:48

Quote:

Originally Posted by wombatman

You're right--I started with minp=1e15 since ppsieve_cuda wouldn't allow anything smaller than that. I will try running FermFact first and just remove the factors found so far by ppsieve_cuda. Then I will continue onward if appropriate.

If so, also consider using FermFact 2.0: it has SSE2 acceleration and a number of different types of file (newpgen, ABC, ABCD...) for the archive.

rogue · 2017-07-13, 16:53

Are you saying that you need to use FermFact, then ppsieve_cuda, then pfgw?

ET_ · 2017-07-13, 17:01

Quote:

Originally Posted by rogue

Are you saying that you need to use FermFact, then ppsieve_cuda, then pfgw?

Not quite.
I am saying that I use FermFact (for 6/9 hours), then ppsieve_cuda, then pfgw.

TBTH, FermFact could be used just to create the starting file, as it reaches the minp value quite soon, but then the file created by FermFact becomes huge (i.e. the file contains millions of candidates) and I prefer letting it run a bit to sieve out the smaller ones.

wombatman · 2017-07-13, 17:13

Quote:

Originally Posted by ET_

If so, also consider using FermFact 2.0: it has SSE2 acceleration and a number of different types of file (newpgen, ABC, ABCD...) for the archive.

Quote:

Originally Posted by ET_

Not quite.
I am saying that I use FermFact (for 6/9 hours), then ppsieve_cuda, then pfgw.

TBTH, FermFact could be used just to create the starting file, as it reaches the minp value quite soon, but then the file created by FermFact becomes huge (i.e. the file contains millions of candidates) and I prefer letting it run a bit to sieve out the smaller ones.

I did grab Version 2.0. I'll use it to get up to something at least close to the minp ppsieve_cuda wants. Thanks for pointing this out to me!

ET_ · 2017-07-13, 17:16

Quote:

Originally Posted by wombatman

I did grab Version 2.0. I'll use it to get up to something at least close to the minp ppsieve_cuda wants. Thanks for pointing this out to me!

You are always welcome, Ben

2017-07-13, 15:17	#46
wombatman I moo ablest echo power! May 2013 73C₁₆ Posts	Gonna take Code: 181744-189999 13000 30000 since I can run both of them at the same time with a relatively minimal (<25%) dropoff in prime throughput. Time per factor for this range will need to be ~25 seconds. Time per factor for the previous range needs to be just ~0.1 sec. Last fiddled with by wombatman on 2017-07-13 at 16:03

2017-07-13, 15:58	#48
wombatman I moo ablest echo power! May 2013 2²×463 Posts	I don't know if it will output a file at the end of a run, but it outputs the factors as Code: 1000000016401549 \| 833778745*2^6246+1 So worst case scenario, you could generate a list, eliminate small factors with srsieve, and then remove the larger factors with srfile. When the shorter run finishes, I'll let you know whether a file with the remaining candidates is generated.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Reservations	ET_	Operazione Doppi Mersennes	539	2023-01-18 00:37
Reservations	kar_bon	Riesel Prime Data Collecting (k*2^n-1)	129	2016-09-05 09:23
Reservations?	R.D. Silverman	NFS@Home	15	2015-11-29 23:18
Reservations	R.D. Silverman	Cunningham Tables	15	2011-03-04 21:01
4-5M Reservations	paulunderwood	3*2^n-1 Search	15	2008-06-08 03:29

2017-07-13, 13:45	#45
wombatman I moo ablest echo power! May 2013 1852₁₀ Posts	Ok. I'll let it run for a while and see how things progress. I'll go ahead and reserve n=6201-6299 for k=8e8-9e8 (the first entry in Rogue's list).

2017-07-13, 15:48	#47
rogue "Mark" Apr 2003 Between here and the 1B33₁₆ Posts	Thanks wombatman. I'm really curious to see how many candidates remain after sieving. BTW, what is the format for the output from ppsieve_cuda? I assume it is the pfgw ABCD format. I will likely need to update PRPNet to support that format.

2017-07-13, 16:53	#52
rogue "Mark" Apr 2003 Between here and the 3×11×211 Posts	Are you saying that you need to use FermFact, then ppsieve_cuda, then pfgw?