CUDALucas not redoing iterations with larger FFT

patrik · 2014-09-11, 14:52

I just had the following problem with CUDALucas v2.04 Beta:

Code:

Iteration 24020000 M( 36163033 )C, 0x47f4b44347a56afb, n = 1920K, CUDALucas v2.04 Beta err = 0.2812 (1:07 real, 6.7233 ms/iter, ETA 22:40:21)
Iteration 24030000 M( 36163033 )C, 0x5f9ad444634e1734, n = 1920K, CUDALucas v2.04 Beta err = 0.2734 (1:08 real, 6.7129 ms/iter, ETA 22:37:07)
Iteration = 24037901 >= 1000 && err = 0.35938 >= 0.35, fft length = 1920K, not writing checkpoint file (because -t is disabled) and exiting.

In this case Prime95 would increase the FFT length and restart from the last checkpoint with a larger FFT size, but CUDALucas just fails at the same point after every restart.

Is there a way I can manually set it to use a larger FFT to get past this point?

(I know this is not the lastest version of CUDALucas, but if I run a later version I use on another computer I just get:
./CUDALucas-2.05-Beta-r64-Linux-x86-64: error while loading shared libraries: libcufft.so.5.5: cannot open shared object file: No such file or directory )

patrik · 2014-09-11, 20:00

I managed to get past the point using the same fft size by changing 0.35 in the source to 0.40, and recompiling the program. (I compiled it from source when I first started using this version.)

I used the new version only for 12651 iterations, then stopped it and switched back to the original version.

In a day or so I will se if this double-check matches the original test.

owftheevil · 2014-09-12, 00:56

If I remember correctly 2.04 can take a command line parameter eg, -f=4000k to use a particular size fft. On the other hand, I don't see any reason that what you have done will give an incorrect result.

2014-09-11, 14:52	#1
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 5²·17 Posts	CUDALucas not redoing iterations with larger FFT I just had the following problem with CUDALucas v2.04 Beta: Code: Iteration 24020000 M( 36163033 )C, 0x47f4b44347a56afb, n = 1920K, CUDALucas v2.04 Beta err = 0.2812 (1:07 real, 6.7233 ms/iter, ETA 22:40:21) Iteration 24030000 M( 36163033 )C, 0x5f9ad444634e1734, n = 1920K, CUDALucas v2.04 Beta err = 0.2734 (1:08 real, 6.7129 ms/iter, ETA 22:37:07) Iteration = 24037901 >= 1000 && err = 0.35938 >= 0.35, fft length = 1920K, not writing checkpoint file (because -t is disabled) and exiting. In this case Prime95 would increase the FFT length and restart from the last checkpoint with a larger FFT size, but CUDALucas just fails at the same point after every restart. Is there a way I can manually set it to use a larger FFT to get past this point? (I know this is not the lastest version of CUDALucas, but if I run a later version I use on another computer I just get: ./CUDALucas-2.05-Beta-r64-Linux-x86-64: error while loading shared libraries: libcufft.so.5.5: cannot open shared object file: No such file or directory )

2014-09-12, 00:56	#3
owftheevil "Carl Darby" Oct 2012 Spring Mountains, Nevada 473₈ Posts	If I remember correctly 2.04 can take a command line parameter eg, -f=4000k to use a particular size fft. On the other hand, I don't see any reason that what you have done will give an incorrect result. Last fiddled with by owftheevil on 2014-09-12 at 00:57

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Coordination thread for redoing P-1 factoring	ixfd64	Lone Mersenne Hunters	80	2021-01-19 17:17
Redoing work?	MooMoo2	Riesel Prime Search	2	2010-09-15 17:33
larger L2 cache, slower iterations?	ixfd64	Hardware	3	2008-05-19 20:46
P-1: poorly P-1'd exponents that need redoing	James Heinrich	Marin's Mersenne-aries	54	2008-04-07 20:27
5.98M to 6.0M: redoing factoring to 62 bits	GP2	Lone Mersenne Hunters	0	2003-11-19 01:30

2014-09-11, 20:00	#2
patrik "Patrik Johansson" Aug 2002 Uppsala, Sweden 5²·17 Posts	I managed to get past the point using the same fft size by changing 0.35 in the source to 0.40, and recompiling the program. (I compiled it from source when I first started using this version.) I used the new version only for 12651 iterations, then stopped it and switched back to the original version. In a day or so I will se if this double-check matches the original test.