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

[storm5510]

Originally Posted by VBCurtis

Please- not every post in this thread is in response to you. You didn't create two new threads about your problem, so it's pretty clear the post you quoted wasn't in response to you nor your problem. Neither was the previous post from Micke, which you replied to- that poster was asking about a new problem, and wasn't in reply to anything you posted.

Many apologies!

Anything quoted not written by me, I will ignore.

[petrw1]

Yesterday I added RAM to one PC.
It went from 8GB to 12GB; 2+2 in 1 channel and 4+4 in the other; all DDR3 1600
Following the project recommendations for B1/B2 from this project I reduced the B1 from 8M to 6.5M for the remaining assignments accordingly.

I did note that any assignments that had already started at 8M continued with 8M; as they should.
However, they got confused in stage 2 .... see 1st exponent below.
Assignments that started with the new 6.5M completed as expected ... 2nd exponent below

. . . worker starting immediately after RAM upgrade.
. . . NOTE this assignment started Stage 1 with B1=8M 
[Jan 3 19:05] P-1 on M4763021 with B1=6500000, B2=TBD
[Jan 3 19:05] Using AVX FFT length 240K, Pass1=320, Pass2=768, clm=2
[Jan 3 19:05] Ignoring suggested B1 value, using B1=8000000 from the save file
[Jan 3 23:31] M4763021 stage 1 complete. 19955792 transforms. Total time: 15993.817 sec.
[Jan 3 23:31] Round off: 0.25
[Jan 3 23:32] Inversion of stage 1 result complete. 5 transforms, 1 modular inverse. Time: 1.527 sec.
[Jan 3 23:32] Exceeded limit on number of workers that can use lots of memory.
[Jan 3 23:32] Looking for work that uses less memory.
. . . SNIP ..... other Stage 1
. . . NOTE the following Stage 2 is missing the "Optimal B2" line and thereby chose a pre-V30.8 B2 value
. . . NOTE also it "forgot" that the B1 was completed above to 8M.
[Jan 4 02:47] 
[Jan 4 03:16] P-1 on M4763021 with B1=6500000, B2=TBD
[Jan 4 03:16] Using AVX FFT length 240K, Pass1=320, Pass2=768, clm=2
[Jan 4 03:16] Available memory is 10228MB.
[Jan 4 03:16] Switching to AVX FFT length 288K, Pass1=384, Pass2=768, clm=2
[Jan 4 03:16] Estimated stage 2 vs. stage 1 runtime ratio: 0.086
[Jan 4 03:16] Using 10227MB of memory.  D: 8190, 864x3726 polynomial multiplication.
[Jan 4 03:17] Stage 2 init complete. 25737 transforms. Time: 79.954 sec.
[Jan 4 03:18] M4763021 stage 2 at B2=76904100 [2.70%]
[Jan 4 03:46] M4763021 stage 2 complete. 598317 transforms. Total time: 1705.091 sec.
[Jan 4 03:46] Round off: 0.125
[Jan 4 03:46] Stage 2 GCD complete. Time: 0.906 sec.
[Jan 4 03:46] M4763021 completed P-1, B1=6500000, B2=665994420, Wi4: 5375CB16
[Jan 4 03:46] 
. . . SNIP
. . . NOTE this assignment started with the new 6.5M B1 and is correctly using the "optimal B2"
[Jan 4 04:16] P-1 on M4763041 with B1=6500000, B2=TBD
[Jan 4 04:16] Using AVX FFT length 240K, Pass1=320, Pass2=768, clm=2
[Jan 4 04:16] Available memory is 10228MB.
[Jan 4 04:16] Switching to AVX FFT length 288K, Pass1=384, Pass2=768, clm=2
[Jan 4 04:16] With trial factoring done to 2^71, optimal B2 is 1291*B1 = 8391500000.
[Jan 4 04:16] If no prior P-1, chance of a new factor is 11.9%
[Jan 4 04:16] Estimated stage 2 vs. stage 1 runtime ratio: 1.277
[Jan 4 04:16] Using 10227MB of memory.  D: 8190, 864x3726 polynomial multiplication.
[Jan 4 04:17] Stage 2 init complete. 25751 transforms. Time: 79.418 sec.
[Jan 4 04:18] M4763041 stage 2 at B2=780539760 [0.21%]

[James Heinrich]

Originally Posted by Prime95

Pfactor= does not support a list of known factors.

Any progress on adding such support?

[Prime95]

Originally Posted by James Heinrich

Any progress on adding such support?

Well maybe, the code says:

/* Starting in 30.9:									*/
/*	Pfactor=k,b,n,c,how_far_factored,ll_tests_saved_if_factor_found[,known_factors]	*/

[harlee]

Originally Posted by Prime95

Well maybe, the code says:

/* Starting in 30.9:									*/
/*	Pfactor=k,b,n,c,how_far_factored,ll_tests_saved_if_factor_found[,known_factors]	*/

Just tested five exponents with 30.10b1 on a MacBook laptop. No factors found in either stage 1 or stage 2.

[preda]

Originally Posted by petrw1

Yesterday I added RAM to one PC.
It went from 8GB to 12GB; 2+2 in 1 channel and 4+4 in the other; all DDR3 1600

A bit off-topic, but you should also consider 4+2 GB in each channel (there may be a benefit when the two channels are balanced).

If you're on an Intel CPU, you can use the mlc tool for measuring the RAM bandwidth:
https://www.intel.com/content/www/us...y-checker.html

[harlee]

Originally Posted by harlee

Just tested five exponents with 30.10b1 on a MacBook laptop. No factors found in either stage 1 or stage 2.

Sorry, can't edit my previous post. Anyway, here is some more information. My MacBook Air is an early 2015 model with 8GB of memory, which 4GB is allocated for P-1 Stage 2 work. I run one worker with one CPU to keep the heat down. Anyway, before going to bed last night, I decided to do 20 more Pfactor tests - 10 in the low 4M range and 10 in the low 5M range. One factor was found. Based on previous P-1 testing with v30.8b15, it would appear the Pfactor is calculating smaller B2 bounds then Pminus1 tests. If interested, I can do more testing to verify this observation but don't know if that would still be true or not with multi-core systems with lots of memory or if it really matters.

M4173683 has a 82.199-bit (25-digit) factor: 5551675757085812597644151 (P-1,B1=312000,B2=166143120)

Since Prime95's note for 30.9 states "If you aren't doing ECM, do not download! PRE-BETA software!", I going back to runnng 30.8b15, unless Prime95 tells me otherwise.

[tha]

I tried a P-1 run on M325133 with B1 set to 1,000,000,000. the machine has 64 GB and normally runs the B2 part using all available RAM. For some reason it entirely skipped the B2 part. This while Stage1GCD=-1.

[Work thread Jan 10 14:37] M325133 stage 1 complete. 1685648344 transforms. Total time: 34128.157 sec.
[Work thread Jan 10 14:37] Stage 1 GCD complete. Time: 0.015 sec.
[Work thread Jan 10 14:37] M325133 completed P-1, B1=1000000000, Wi8: 7FBE989B
[Comm thread Jan 10 14:37] Sending result to server: UID: Tha/Z-170, M325133 completed P-1, B1=1000000000, Wi8: 7FBE989B, AID: 2F1763106C011E5B744B722A29FDCB09

[Prime95]

Originally Posted by tha

I tried a P-1 run on M325133 with B1 set to 1,000,000,000. the machine has 64 GB and normally runs the B2 part using all available RAM. For some reason it entirely skipped the B2 part. This while Stage1GCD=-1.

I'd have to see the worktodo.txt entry to decide if this is a bug

[tha]

Originally Posted by Prime95

I'd have to see the worktodo.txt entry to decide if this is a bug

I believe this has been the worktodo line:

PMinus1=1,2,325133,-1,1000000000,40000000,68

The program was restarted after the line was added to the worktodo file. The server reserved the exponent and gave an assignment key. I considered the B2 value as irrelevant since it is supposed to be ignored. These are the first lines of the run:

[Comm thread Jan 9 19:07] Registering assignment: P-1 M325133
[Comm thread Jan 9 19:07] Assignment registered as: 2F1763106C011E5B744B722A29FDCB09
[Comm thread Jan 9 19:07] Sending expected completion date for M325133: Jan 10 2023
[Comm thread Jan 9 19:07] Done communicating with server.
.
.
.
[Work thread Jan 9 19:23] 
[Work thread Jan 9 19:23] P-1 on M325133 with B1=1000000000
[Work thread Jan 9 19:23] Setting affinity to run helper thread 1 on CPU core #2
[Work thread Jan 9 19:23] Using FMA3 FFT length 16K, Pass1=256, Pass2=64, clm=2, 4 threads
[Work thread Jan 9 19:23] Setting affinity to run helper thread 3 on CPU core #4
[Work thread Jan 9 19:23] Setting affinity to run helper thread 2 on CPU core #3

Linux64,Prime95,v30.8,build 15

[James Heinrich]

I would suspect the logic is something like
if B2==0 { auto-set B2 }
if B2 > B1 { do stage 2 }
Since neither of those are true, it did stage 1 only