Prime95 30.8 (big P-1 changes, see post #551) - Page 11

Prime95 · 2021-12-06, 06:52

Quote:

Originally Posted by ATH

Switched from build2 to build3 during Stage1. When it reaches P-1 Stage2 it freezes during initialization

Preliminary analysis is a deadlocking issue. Try running stage 2 single threaded.

lycorn · 2021-12-06, 08:09

Prime95 (30.8 build 3) silently stopped working at the beginning of P-1 stage 2 - last messages were about setting the affinities for the 3 helper threads (it´s a 4 core machine). Several hours later, Task Manager was showing 0% CPU and 2.3 GB of RAM in use - it is allowed to use up to 11 GB, which it normally does). The green icon was visible in the tray, and right clicking it produced the usual menu; selecting stop would display the message "stopping all working windows" in the comms window, but nothing actually happened. I had to use Task Manager to kill the process.
The machine had been running build 3 for many hours without any problem, and had already tested several exponents
Am I the only one to experience this problem?

I just noticed ATH posted a similar situation. In my case I had been running build 3 from the beginning of stage 1, tho.

Prime95 · 2021-12-06, 10:10

Quote:

Originally Posted by lycorn

I just noticed ATH posted a similar situation. In my case I had been running build 3 from the beginning of stage 1, tho.

Quote:

Originally Posted by Prime95

Preliminary analysis is a deadlocking issue. Try running stage 2 single threaded.

Found and fixed the deadlock. The smaller the exponent you are testing (or the more threads you are running) the more likely you are to experience a deadlock. I'll post a fix as soon as I can.

Xyzzy · 2021-12-06, 13:51

axn · 2021-12-06, 14:17

Nice! A condensed summary of the run:

Code:

Optimal P-1 factoring of M108028051 using up to 231424MB of memory.
Assuming no factors below 2^76 and 1.1 primality tests saved if a factor is found.
Optimal bounds are B1=755000, B2=479779000
Chance of finding a factor is an estimated 6.32%

Using AVX-512 FFT length 5760K, Pass1=3K, Pass2=1920, clm=1, 18 threads
M108028051 stage 1 complete. 2178208 transforms. Total time: 2516.060 sec.
Conversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 45.094 sec.
Switching to AVX-512 FFT length 7056K, Pass1=1344, Pass2=5376, clm=2, 18 threads
Using 229108MB of memory. D: 6930, 720x3371 polynomial multiplication.
Stage 2 init complete. 20328 transforms. Time: 211.952 sec.
M108028051 stage 2 complete. 721861 transforms. Total time: 2481.481 sec.
Stage 2 GCD complete. Time: 24.941 sec.
M108028051 completed P-1, B1=755000, B2=492127020

With huge amount of RAM, this method could potentially speed up PRP wavefront by 2%

alpertron · 2021-12-06, 16:25

Assuming 10% of the users have huge amounts of RAM, that will translate to only 0.2%

VBCurtis · 2021-12-06, 16:40

Quote:

Originally Posted by alpertron

Assuming 10% of the users have huge amounts of RAM, that will translate to only 0.2%

No, because even 5% of users is plenty to P-1 all the wavefront exponents. Axn is saying 2% more factors would be found, saving that many prp tests and speeding the project by 2%.

James Heinrich · 2021-12-06, 16:48

Just playing around with 30.8b3.win64 (on M859433, yes I know it's prime) and found some odd behaviours. i7-3930K, 64GB (32GB allocated).
Tried with the default 6 threads and it froze as described above. Limited it to 1 thread and it worked:

Quote:

[Dec 6 10:23] Worker starting
[Dec 6 10:23] Setting affinity to run worker on CPU core #2
[Dec 6 10:23] Optimal P-1 factoring of M859433 using up to 32768MB of memory.
[Dec 6 10:23] Assuming no factors below 2^0 and 10 primality tests saved if a factor is found.
[Dec 6 10:23] Optimal bounds are B1=1000, B2=104000
[Dec 6 10:23] Chance of finding a factor is an estimated 100%
[Dec 6 10:23]
[Dec 6 10:23] Using AVX FFT length 48K, Pass1=256, Pass2=192, clm=1
[Dec 6 10:23] M859433 stage 1 complete. 3040 transforms. Total time: 0.601 sec.
[Dec 6 10:24] Conversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 0.422 sec.
[Dec 6 10:24] Using 32768MB of memory. D: 19110, 2016x85104 polynomial multiplication.
[Dec 6 10:24] Round off: 0.00439453125, poly_size: 2, EB: 6.24679, SM: 0
[Dec 6 10:24] Round off: 0, poly_size: 4
[Dec 6 10:24] Round off: 0, poly_size: 8
[Dec 6 10:24] Round off: 0, poly_size: 16
[Dec 6 10:24] Round off: 0, poly_size: 32
[Dec 6 10:24] Round off: 0, poly_size: 64
[Dec 6 10:24] Round off: 0, poly_size: 128
[Dec 6 10:24] Round off: 0, poly_size: 256
[Dec 6 10:24] Round off: 0, poly_size: 512
[Dec 6 10:24] Round off: 0, poly_size: 1024
[Dec 6 10:24] Round off: 0, poly_size: 2048
[Dec 6 10:24] Stage 2 init complete. 61794 transforms. Time: 54.750 sec.
[Dec 6 10:29] Round off: 0
[Dec 6 10:30] M859433 stage 2 complete. 666722 transforms. Total time: 343.713 sec.
[Dec 6 10:30] Starting stage 2 GCD - please be patient.
[Dec 6 10:30] Stage 2 GCD complete. Time: 0.180 sec.
[Dec 6 10:30] M859433 completed P-1, B1=1000, B2=1549324140, Wi4: 4A39F218
[Dec 6 10:30] No work to do at the present time. Waiting.

Optimal bounds are described as 104:1, but then it proceeds to select a B2 that 1.5million times bigger than B1, I'm not sure that's intentional? And B2:B1 runtime is 336x.
I'm not sure the "100% chance of factor" is quite right either.

axn · 2021-12-06, 16:59

I think you may have messed up the worktodo line. To wit:

Quote:

Assuming no factors below 2^0

What was the worktodo line you used?

Yes, it is normal for it to use larger than the specified B2.

alpertron · 2021-12-06, 17:45

Quote:

Originally Posted by VBCurtis

No, because even 5% of users is plenty to P-1 all the wavefront exponents. Axn is saying 2% more factors would be found, saving that many prp tests and speeding the project by 2%.

We do not know if these people will accept doing only P-1. Some people only wants to perform PRP because of the prizes. So they would use multithreading to run as many PRP as possible.

James Heinrich · 2021-12-06, 17:46

Quote:

Originally Posted by axn

What was the worktodo line you used?

Pfactor=1,2,859433,-1,0,10 was what I used for the above. (I came across this exponent on my "worst P-1" page, before I excluded prime exponents, and it had the erroneous TF=0 level). Perhaps Prime95 should apply some sanity checks to specified TF levels?
Running it again with Pfactor=1,2,859433,-1,40,10 and the numbers come up more sane:

Quote:

[Dec 6 12:30] Optimal P-1 factoring of M859433 using up to 32768MB of memory.
[Dec 6 12:30] Assuming no factors below 2^40 and 10 primality tests saved if a factor is found.
[Dec 6 12:30] Optimal bounds are B1=82000, B2=670408000
[Dec 6 12:30] Chance of finding a factor is an estimated 42%
[Dec 6 12:30] Using AVX FFT length 48K, Pass1=256, Pass2=192, clm=1
[Dec 6 12:31] M859433 stage 1 complete. 236894 transforms. Total time: 61.151 sec.
[Dec 6 12:31] Conversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 0.248 sec.
[Dec 6 12:31] Switching to AVX FFT length 50K, Pass1=640, Pass2=80, clm=1
[Dec 6 12:31] Using 32440MB of memory. D: 90090, 8640x74168 polynomial multiplication.
[Dec 6 12:32] Round off: 0, poly_size: 2, EB: 0.581153, SM: 3.5192
[Dec 6 12:32] Round off: 0, poly_size: 4
[Dec 6 12:32] Round off: 0, poly_size: 8
[Dec 6 12:32] Round off: 0, poly_size: 16
[Dec 6 12:33] Round off: 0, poly_size: 32
[Dec 6 12:33] Round off: 0, poly_size: 64
[Dec 6 12:33] Round off: 0, poly_size: 128
[Dec 6 12:33] Round off: 0, poly_size: 256
[Dec 6 12:34] Round off: 0, poly_size: 512
[Dec 6 12:34] Round off: 0, poly_size: 1024
[Dec 6 12:35] Round off: 0, poly_size: 2048
[Dec 6 12:35] Round off: 0, poly_size: 4096
[Dec 6 12:35] Round off: 0, poly_size: 8192
[Dec 6 12:36] Round off: 0, poly_size: 16384
[Dec 6 12:36] Stage 2 init complete. 313970 transforms. Time: 281.277 sec.
[Dec 6 12:39] Round off: 0
[Dec 6 12:39] M859433 stage 2 complete. 532866 transforms. Total time: 183.998 sec.
[Dec 6 12:39] Starting stage 2 GCD - please be patient.
[Dec 6 12:39] Stage 2 GCD complete. Time: 0.097 sec.
[Dec 6 12:39] M859433 completed P-1, B1=82000, B2=5164589430, Wi4: 25B7808A

Estimated B2:B1 prior to start was 8000:1, ended up with 63000:1 which I think makes more sense than 1500000:1, as well as the stage2:1 time ratio of ~8:1 is better than 330:1.

One side oddity:
Sending result to server: UID: JamesHeinrich, M859433 completed P-1, B1=82000, B2=5164589430, Wi4: 25B7808A
But the result shows up as "-Anonymous-".

2021-12-06, 08:09	#112
lycorn "GIMFS" Sep 2002 Oeiras, Portugal 62E₁₆ Posts	Prime95 (30.8 build 3) silently stopped working at the beginning of P-1 stage 2 - last messages were about setting the affinities for the 3 helper threads (it´s a 4 core machine). Several hours later, Task Manager was showing 0% CPU and 2.3 GB of RAM in use - it is allowed to use up to 11 GB, which it normally does). The green icon was visible in the tray, and right clicking it produced the usual menu; selecting stop would display the message "stopping all working windows" in the comms window, but nothing actually happened. I had to use Task Manager to kill the process. The machine had been running build 3 for many hours without any problem, and had already tested several exponents Am I the only one to experience this problem? I just noticed ATH posted a similar situation. In my case I had been running build 3 from the beginning of stage 1, tho. Last fiddled with by lycorn on 2021-12-06 at 08:11 Reason: Cross posting

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Do not post your results here!	kar_bon	Prime Wiki	40	2022-04-03 19:05
what should I post ?	science_man_88	science_man_88	24	2018-10-19 23:00
Where to post job ad?	xilman	Linux	2	2010-12-15 16:39
Moderated Post	kar_bon	Forum Feedback	3	2010-09-28 08:01
Something that I just had to post/buy	dave_0273	Lounge	1	2005-02-27 18:36

2021-12-06, 13:51	#114
Xyzzy Aug 2002 2·3²·13·37 Posts

2021-12-06, 16:25	#116
alpertron Aug 2002 Buenos Aires, Argentina 1,523 Posts	Assuming 10% of the users have huge amounts of RAM, that will translate to only 0.2%