genefer/CUDA - Page 15

prime7989 · 2012-10-12, 02:20

I tried genefercuda for a gtx680 nvidia GPU on a AMD FX8120 cpu and 32 gb ram for
Fermat number F19 which is known composite with b=2 and N=524288=2^19
and the program prints out that it is probably prime.
Can any one else try this out?
A...

Batalov · 2012-10-12, 02:47

Quote:

Originally Posted by prime7989

I tried genefercuda for a gtx680 nvidia GPU on a AMD FX8120 cpu and 32 gb ram for
Fermat number F19 which is known composite with b=2 and N=524288=2^19
and the program prints out that it is probably prime.
Can any one else try this out?
A...

genefercuda is running a PRP test in base 2 (just checked the source), which is fine for checking any GF(b), b>2 (and not a power of 2, which would be pointless).
Of course, all Fermat numbers are 2-PRPs.

axn · 2012-10-12, 03:35

Quote:

Originally Posted by Batalov

genefercuda is running a PRP test in base 2 (just checked the source), which is fine for checking any GF(b), b>2 (and not a power of 2, which would be pointless).
Of course, all Fermat numbers are 2-PRPs.

It should be relatively trivial to change the PRP-base (to say, 3). I will try it out later.

Also, 2^524288 = 65536^32768. So a much smaller FFT may be used.

axn · 2012-10-12, 13:52

Quote:

Originally Posted by axn

It should be relatively trivial to change the PRP-base (to say, 3). I will try it out later.

Also, 2^524288 = 65536^32768. So a much smaller FFT may be used.

So, I tried this out... and you wouldn't believe what I found out:

Code:

65536^32768+1 is composite. (RES=0df151a4161b6c99) (157827 digits) (err = 0.0005) (time = 0:07:24)

LaurV · 2012-10-13, 12:19

That's creepy! Don't say all the tests need to be redone?

msft · 2012-10-15, 01:26

Quote:

Originally Posted by LaurV

That's creepy! Don't say all the tests need to be redone?

I Bet your side.

LaurV · 2012-10-15, 09:12

Oh, I forgot about this, my post was kinda joke, based on the fact that the incriminated number is F19, which is known as composite. If axn did a 3-PRP test it was normal to be composite (it is like a Pepin Test.... squared, I might be wrong here but this will never return a false result, i.e. Fx is 3-prp if and only if it is prime).

msft · 2012-10-16, 06:35

base=3

Code:

$ diff genefer.c genefer.c.ORG
1757c1757
<       CONST double t1 = 2.0 / m, t2 = 3 * t1;
---
>       CONST double t1 = 2.0 / m, t2 = 2 * t1;
1781c1781
<       z = FFTinitGFN(m, 3.0, &n1, &n2);
---
>       z = FFTinitGFN(m, 2.0, &n1, &n2);
$ cc genefer.c -lm
$ ./a.out -t
GeneFer 1.3 (standard)  Copyright (C) 2001-2002, Yves GALLOT
A program for finding large probable generalized Fermat primes.

Usage: GeneFer -b            run bench
       GeneFer -t            run test
       GeneFer <filename>    test <filename>
       GeneFer               use interactive mode

100050^16+1 is a probable prime. (err = 1.53e-05)
100014^32+1 is a probable prime. (err = 3.05e-05)
100234^64+1 is a probable prime. (err = 4.58e-05)
100032^128+1 is a probable prime. (err = 9.16e-05)
5684328^256+1 is a probable composite (err = 5.00e-01).
4619000^512+1 is a probable composite (err = 5.00e-01).
3752220^1024+1 is a probable composite (err = 5.00e-01).

Batalov · 2012-10-16, 06:45

Aaah! msft is Morra!
(...I suddenly expect Morra defence in the Vote chess thread. No. I did expect it before. That's when I looked up Morra.
Too late. ;-) )

"The Groke" on the other hand reminds me of Piet Hein's Grooks.

Quote:

THE ROAD TO WISDOM?

Well, it's plain
and simple to express.
Err and err and err again,
but less and less and less.

prime7989 · 2013-03-17, 04:54

Quote:

Originally Posted by axn

So, I tried this out... and you wouldn't believe what I found out:

Code:

65536^32768+1 is composite. (RES=0df151a4161b6c99) (157827 digits) (err = 0.0005) (time = 0:07:24)

Dear axn,
I dislike reading code. So please which variable did you change to make it 3-PRP
in GeneferCUDA.cu, please? If you could give me the line number of the code of function and var, I would be very greatful!
Thank you,
prime7989

axn · 2013-03-17, 08:23

Quote:

Originally Posted by prime7989

Dear axn,
I dislike reading code. So please which variable did you change to make it 3-PRP
in GeneferCUDA.cu, please? If you could give me the line number of the code of function and var, I would be very greatful!
Thank you,
prime7989

There are many versions of G/CUDA out there. If you can post the one you want to look at (as a forum attachment), I can tell you. FWIW, the change is easy and localised to "check" function.

2012-10-12, 02:20	#155
prime7989 Jun 2012 17 Posts	Is genefercuda correct I tried genefercuda for a gtx680 nvidia GPU on a AMD FX8120 cpu and 32 gb ram for Fermat number F19 which is known composite with b=2 and N=524288=2^19 and the program prints out that it is probably prime. Can any one else try this out? A...

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Genefer's FFT applied to Mersenne squaring	preda	Software	0	2017-09-06 02:54
CUDA 5.5	ET_	GPU Computing	2	2013-06-13 15:50
AVX CPU LL vs CUDA LL	nucleon	GPU Computing	11	2012-01-04 17:52
Best CUDA GPU for the $$	Christenson	GPU Computing	24	2011-05-01 00:06
CUDA?	Xentar	Conjectures 'R Us	6	2010-03-31 07:43

2012-10-13, 12:19	#159
LaurV Romulan Interpreter Jun 2011 Thailand 3×3,221 Posts	That's creepy! Don't say all the tests need to be redone?

2012-10-15, 09:12	#161
LaurV Romulan Interpreter Jun 2011 Thailand 3·3,221 Posts	Oh, I forgot about this, my post was kinda joke, based on the fact that the incriminated number is F19, which is known as composite. If axn did a 3-PRP test it was normal to be composite (it is like a Pepin Test.... squared, I might be wrong here but this will never return a false result, i.e. Fx is 3-prp if and only if it is prime).