mersenneforum.org How to read the "Exponent Status Distribution"
 Register FAQ Search Today's Posts Mark Forums Read

2020-07-24, 10:03   #23
LaurV
Romulan Interpreter

Jun 2011
Thailand

221E16 Posts

Quote:
 Originally Posted by Uncwilly The blue line is the 2-3 bits more than that that the average GPU should do. The green line is the 5 bits higher that Ken talked about.
Nitpicking, to be on the correct side, the green line is what an average GPU should do.

The blue line immediately under it, is what an average GPU should do when TF-ing at DC front, i.e. exponents that already had one LL or PRP test done.

The difference is exactly 1 bitlevel, from objective reasons: finding a factor will save 1 LL/PRP test in case of DC, but will save 2 tests in case of first-time LL/PRP (i.e. the exponents had no test done). So the amount of saved work is double at the FT front line, compared to DC front line, but on the other hand, there is a double amount of TF candidates between 2^n and 2^(n+1) compared to the amount of candidates between 2^(n-1) and 2^n, so the effort to go one bitlevel higher is double. Assuming the probability of finding a factor is the same at the two bitlevels (which is somehow true, in a reasonable range of error) then you should go 1 bitlevel higher with your TF when you TF at the FT front line.

Last fiddled with by LaurV on 2020-07-24 at 10:04

 2020-07-24, 13:59 #24 kriesel     "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 113816 Posts And, as PRP with proof and verification rolls out, the 2 tests vs 1 saved increasingly becomes 1 and a small fraction if no tests have been done yet. A 97M PRP test is ~300+GhzD, but a CERT for one may be only 1.4GhzD at power 8. Recent Gpuowl supports proof and verification; prime95 V30.1b1 supports proof and verification.

 Similar Threads Thread Thread Starter Forum Replies Last Post heliosh Information & Answers 6 2020-07-20 19:27 pinhodecarlos NFS@Home 2 2015-07-04 11:18 James Heinrich Data 2 2012-02-01 21:14 ixfd64 Lounge 3 2004-05-27 00:51 ixfd64 Lounge 0 2003-10-14 23:04

All times are UTC. The time now is 08:59.

Sun Sep 20 08:59:02 UTC 2020 up 10 days, 6:10, 0 users, load averages: 1.24, 1.45, 1.47