Some things can be done more efficiently if they are "vectorized" but not all the algorithms can be easily vectorized with the current technology and maybe that chess games belong to this case, I'm not expert about them.
For what concerns our case, there's a lot of things that we can do in order to make steps that we could not afford in the past but that the technology we already own in some way now allows. If we had incredibly fast computers we could simply decode a message voting for the most likely among all the possible messages that could have been transmitted but this, on actual computers, is still infeasible even with the quite short messages we use on WSJTX. So far, if we would get out the best decoding results given the available processing capabilities and assuming that we don't want to wait years to see if something gets decoded, the way we write the code and exploit some hardware, if available, matters indeed. Getting the best out of the codes used in WSJTX is not just a matter of telecommunication theory. We try to do something that industry likely would never do as it would be not that convenient. Anyway we aim to go beyond compromises and this requires quite an huge processing power that we can possibly exploit from the most modern machines. None of us wants to put pressure on this aspect. I think that Bill's poll was just aimed to see how we could improve what was already done given the hardware that is commonly used. I think that the first goal that WSJTX developers address is still (and has always been) that of thinking and deploying something smart, solid and useful for the community, not that of embroidering funny solutions. Yet the problem of doing what we do in the best way we can, given all the world and the technology that revolves around us, still persists despite we are not explicitly called to cope with it. BTW, I don't see any problem for EMErs. If you already do EME activity a 30% more CPU power consumption (if it should be really required) is just a small fraction of the power used by the transmitter PA. Therefore even a single dB of coding gain would easily translate to a total energy saving. 73 Nico / IV3NWV Il giorno mer 9 giu 2021 alle ore 19:23 Black Michael via wsjt-devel < wsjt-devel@lists.sourceforge.net> ha scritto: > Is there some reasons to think we get a notable performance boost from AVX? > > First link I found is not impressive....very few benchmarks show much > improvement and many show less and it's at a cost of 30% or so more CPU > power consumption...fans will be screaming...looks like it mainly helps on > decoding streams. > > > https://www.phoronix.com/scan.php?page=article&item=rocket-lake-avx512&num=2 > > Mike W9MDB > _______________________________________________ > wsjt-devel mailing list > wsjt-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/wsjt-devel >
_______________________________________________ wsjt-devel mailing list wsjt-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/wsjt-devel
