BOINC says they are unable to determine if the code is invalid. Science United uses that list to assign work. I have had to struggle to figure out the above! Who knows what other errors exist?!ĭare I mention that, as many of us know, the project gives a HUGE amount of BOINC credit! Dare I mention the lack of organized authoritative results? I have emailed BOINC to have them consider removing this project from their published list. I am even more worried because Jon Sonntag, the project admin, refuses to release *any* amount of source code for review. It is important that EVERY number be tested, and I have zero confidence that this is true. (3) AMD and Intel GPUs are known to commonly make arithmetic errors for the 64-bit integers (ulong) used in the code This would especially be a problem if global_work_size is large! (2) maxStep = 2^28-1, which on the many GPUs that have a watchdog timer, may not detect the infinite cycles that would disprove the Collatz conjecture. (1) most devices at this project do not have ECC RAM The following are also all very worrisome, especially considering that, in this project, tasks are not validated by another device. It only shows numbers of pattern 3, 7, 11, 15, 19. You can easily confirm for yourself that this project is still invalid by looking at. I suppose the 1 step of number 2 isn't going to be found either! This sieve will only run 25% of the numbers.īut notice that 8 steps occurs at 6, which would never be tested with this sieve. Let's look at a 2^2 sieve for a very simple example of what this project is doing wrong. This is assuming that this BOINC project's code is valid in any form, and I have many serious concerns from the small amount of code that I can actually see. In light of this, there are algorithms that don't find high steps to reduce a number to 1 that are 78 times faster for CPU (7700% faster) and 36 times faster on GPU (3500% faster), so this BOINC project performs 1% as fast on CPU and 3% as fast on GPU as it should. Please see this paper for a valid algorithm.Īll results of this BOINC Collatz project regarding high steps are invalid. See the kernel code from this project below, and you'll see that a sieve is used. However, you may not use a sieve when doing this. The project has the goal of finding the high steps to reduce a number to 1. Version 1.2 or any later version published by the Free Software Foundation.Bad news everyone, results of the BOINC Collatz project are invalid. Under the terms of the GNU Free Documentation License, Permission is granted to copy, distribute and/or modify this document Also in the original Italian.Ĭopyright © 2023 University of California. Video and paper from The Ramanujan MachineĬheck out a new video and paper from The Ramanujan Machine, which has used BOINC to discover new expressions for fundamental mathematical constants.Ĭheck out The democratization of science: Analysis of the voluntary distributedĬomputing platform BOINC, a Masters thesis by Antonio Cerrato. Thanks to everyone who contributed to developing and testing this release. The 7.24.1 version of the BOINC client software has been released for Windows, Mac, and Android. The prime 41*684^436354+1 has 1.237.090 digits and entered the TOP5000 in Chris Caldwell's The Largest Known Primes Database. Ralfy, a member of the team BOINC Confederation found a megaprime for base S684. Has 660953486 orthogonal mates which puts it on the 16th place of the rating. Processing of the square # 25 completedĭear participants, processing of the square # 25 is fully completed! The square: Thank you for project attention, support and donation of CPU time! Now "high part" of spectra of ODLS-12 looks like this (square # 26 marked by red): Has 532875410 orthogonal mates which puts it on the 18th place of the rating which include now 5445 positions. Processing of the square # 26 completedĭear participants, processing of the square # 26 is fully completed! The square:
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |