20210601, 02:48  #1 
"University student"
May 2021
Beijing, China
103_{10} Posts 
Can we begin PRP using part of the results of P1 stage 1?
As far as I know, in P1 stage 1 we compute 3^{2*p*E} modulo M_{p}. However, in PRP we also compute 3^{2^p} modulo M_{p}. In both cases the base number is 3.
Could we begin PRP using part of the results in P1 (say, 3^{2^k} where k is the highest bit level of 2*p*E) , to save some work (k iterations, about 1%), since most PRP firsttime tests are assigned together with a prior P1 ? I'm a freshman here and I don't know if the idea is feasible. Last fiddled with by Zhangrc on 20210601 at 02:57 
20210601, 05:02  #2 
P90 years forever!
Aug 2002
Yeehaw, FL
2^{4}·3^{2}·53 Posts 
A good idea! I won't go into the details of how one would do such an optimization.
Gpuowl already does this. It is on my list of things to look into for prime95. The one problem is the optimization takes a lot of memory. Thus, for many users there may be little benefit. 
20210601, 10:13  #3  
"University student"
May 2021
Beijing, China
103 Posts 
Quote:
By the way, how much memory will it take? I'm sparing 11GB of memory out of 16GB to run Prime95, is that enough? Also I'm looking forward to see version 30.7 release. 

20210601, 10:35  #4 
Jun 2003
19·271 Posts 
This thread has some related discussions, I believe.

20210601, 11:22  #5 
Jun 2003
19·271 Posts 
And this followup thread as well

20210601, 12:10  #6 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
5,783 Posts 
Estimating space, for each power needed, 64 bits/double / (~18 bits/ fft word) x ceil (p/8). Storage could be in the compressed form, ceil(p/8). For p up to ~10^{9}, log2(10^{9}) ~30. 30 ceil (10^{9}/8) = 3.75GB. The representation as doubles is ~13.3GB, without misc. overhead. That's in addition to the P1 stage 1 requirements. In low ram situations or to save at stop, resume later, it can be spilled to disk and retrieved later.

20210602, 08:17  #7  
"David Kirkby"
Jan 2021
Althorne, Essex, UK
447_{10} Posts 
Quote:
BTW, during stage 2 of P1 factoring of M104212883, which was previously trial factored to 2^{76}, mprime used 303 GB RAM with B1=881,000, B2=52,281,000. Code:
[Worker #2 May 31 12:25] M104212883 stage 1 complete. 2543108 transforms. Time: 1935.398 sec. [Worker #2 May 31 12:25] Starting stage 1 GCD  please be patient. [Worker #2 May 31 12:26] Stage 1 GCD complete. Time: 45.404 sec. [Worker #2 May 31 12:26] Available memory is 376728MB. [Worker #2 May 31 12:26] D: 2730, relative primes: 7088, stage 2 primes: 3059717, pair%=95.64 [Worker #2 May 31 12:26] Using 310498MB of memory. Code:
[Worker #2 May 31 11:53] Assuming no factors below 2^76 and 2 primality tests saved if a factor is found. Last fiddled with by drkirkby on 20210602 at 08:24 

20210602, 09:19  #8 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
5,783 Posts 
Key word there is highlighted in red, a holdover from the days of all LL, no PRP. Since most first testing is now PRP, perhaps a future release will adjust the assumption from 2 to ~1 for first tests, or at least PRP first tests, which would adjust the bounds selection somewhat, for greater overall efficiency of searching for new Mersenne primes.

20210605, 03:13  #9 
"University student"
May 2021
Beijing, China
67_{16} Posts 
Wow!
By the way, shall we use slightly smaller TF bounds, such as 2^75, for exponents of ~115M? I think it's not a must given so much GPU computing power. 
20210605, 06:21  #10  
Jun 2003
19×271 Posts 
Quote:
Current wavefront is around 110m which uses 6M FFT. Given memory of 1GB, we can get 1024/48 = 21 temps. Let's assume we're targeting B1=1.2m which is about 1.73mbits of straight P1. With 16 temps (largest power of 2 < 21), we can do the P1 stage 1 with an additional ~290k multiplies. However, we aren't limited to power of two temps (though that is the easiest to conceptualize). We utilize all 21 temps (handling all 5 bit patterns and a few 6 bit patterns), this reduces effort to ~275k multiplies). Compare this with the optimal 2^13 temps which gets this done in ~132k multiplies. But also compare this with straight P1 which gets this done in 1.73m multiplies!! Also, there is a downside to using a large number of temps. All the temps are part of your state! So if you need to write a checkpoint, you will need to write all of them to the disk. In the optimal case, that is ~110GB of IO per checkpoint. Obviously, this is not good. You could reduce it by accumulating the results and just writing that out  but that would mean 2*temp muls before each checkpoint. Here also, having fewer temps is helpful. In short, less memory is still a major gain, and might be a blessing in disguise. 

20210615, 18:25  #11 
Mar 2004
3·181 Posts 
To piggyback on this idea, could we also go the other way? Could we use a saved value of B1 and also a small power of 3 times one of the last few iterations of the PRP test do a quick 2nd B1 test? Meaning, the following:
Given Mp and iteration m near the end of the PRP test (say on the last day), so m is somewhat close to but below p1. On iteration m, we have found 3^2^m mod Mp. We then do a quick P1 test on 3^[(2^m+k)*B1] mod Mp for 1<=k<=128. Is that feasible? Each 2^m+k should have about 18.4 distinct prime factors. Perhaps 18.4 prime factors for each value of k is enough to make it matter? Or would the GCD stage take too long? Last fiddled with by JuanTutors on 20210615 at 18:25 
Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
How to make GMPECM skip stage 1 when feeding the prime95 results  Ensigm  Factoring  9  20200824 17:32 
When did PrimeNet begin rubberstamping results?  Chuck  PrimeNet  64  20140106 01:22 
Only submit part of ECM results?  dabaichi  PrimeNet  5  20111207 19:27 
In which country do you suggest me to begin my act  Unregistered  Information & Answers  0  20101130 23:34 
When did Rock'n'Roll begin?  davieddy  Lounge  9  20090825 20:01 