20200428, 10:52  #34  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
3×1,579 Posts 
Quote:
That's covered in points 8 and 9 of https://www.mersenneforum.org/showpo...23&postcount=6 (my attempt to document all the concepts of mfaktc trial factoring I could identify, which had been reviewed by the author of mfaktc) But suppose one in 10^{6} primes were omitted somehow from the TF candidate factors. We would still get nearly all the benefits of a perfect TF process, with all or nearly all the run time of a perfect TF process, over the same bit size interval. A usable proof of work algorithm needs to be tolerant of both some level of expected software imperfection and speed optimizations. Last fiddled with by kriesel on 20200428 at 11:01 

20200428, 13:33  #35 
P90 years forever!
Aug 2002
Yeehaw, FL
15767_{8} Posts 
All prime factors will be tested. A small nondeterministic set of composite factors will also be tested.

20200428, 13:52  #36  
If I May
"Chris Halsall"
Sep 2002
Barbados
246E_{16} Posts 
Quote:
Edit: Sorry, I just realized this would be useless. Different for every run. Never mind... Edit2: What about a sum/hash of the vector of candidates /before/ the sieve? Is that deterministic and reproducable? (Sorry... Haven't had my second cup of coffee yet...) Last fiddled with by chalsall on 20200428 at 13:55 

20200428, 15:55  #37  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
3×1,579 Posts 
Quote:
Is this nondeterministic set of composite factors a problem for Gerbicz' TF proof of work method? https://www.mersenneforum.org/showpo...7&postcount=30 If for the set of q=2kp+1 that are composite but are not sieved out, one or more of (Mp mod q) mod 2^{t} is 1, it would seem so. Testing in that case that a candidate k for the proof list gives a prime q (finishing the sieving of only those that are candidates to go into the verification list) should handle that. Maybe that's best done on the cpu. The value of the list formed, as a proof of work, is that it is hard to produce, and easy to verify, rather like a found factor. Also it is usable in the most common TF case, no factor found. And depends on doing most of the actual TF work. And can be made reliable yet somewhat compact in size. And does not depend on keeping secrets such as encryption codes secret. One of the reasons I like the Gerbicz double mod method is something analogous might be usable for P1 verification of work, which is also not currently present. Some fraction of the code could be reused. Last fiddled with by kriesel on 20200428 at 16:25 

20200428, 18:03  #38  
If I May
"Chris Halsall"
Sep 2002
Barbados
2×4,663 Posts 
Quote:
Probably wouldn't help determine sanity during runtime, but it would open the option of random DCs (for kit sanity checks, for example, if the user wanted to check and monitor their own kit). 

20200428, 18:28  #39  
"Seth"
Apr 2019
2×101 Posts 
Quote:
Quote:


20200428, 18:52  #40  
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
3×1,579 Posts 
Quote:


20200428, 23:57  #41  
"Seth"
Apr 2019
2·101 Posts 
Quote:
https://www.mersenneforum.org/showthread.php?t=24994 

20200429, 00:05  #42  
"Seth"
Apr 2019
CA_{16} Posts 
Quote:
A suspect some small percent of TF results were never run because manually created NO_FACTOR results including a larger bitrange were submitted (e.g. a NF result for bitrange 180). Making that harder (via a checksum for the line even) seems like a good starting place. If you want to do more to verify the calculating (to prevent against overclocking / math errors) we should look at my proofofwork proposal. 

20200429, 16:39  #43 
If I May
"Chris Halsall"
Sep 2002
Barbados
2×4,663 Posts 
My problem is keeping the kittens off the bed...

20200429, 16:42  #44 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest
3·1,579 Posts 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
GPU Trial Factoring FAQ  garo  GPU Computing  100  20190422 10:58 
mfaktc for dummies  NBtarheel_33  GPU Computing  10  20111013 00:04 
How much Trial Factoring to do?  odin  Software  4  20100808 20:23 
How far to do trial factoring  S485122  PrimeNet  1  20070906 00:52 
trial factoring and P1  jocelynl  Math  8  20060201 14:12 