asciilifeform: nao it ~is~ possible that in avg case ( again, if not obv, the above could easily be a fluke, given as whole algo is monte carlo ) there is greater economy.
asciilifeform: spoiler : economy not significant (in this ~one~, they are effectively one, rngistically) test , eight 2048bit 'prob primes' shat in 4m26.812s vs 3m52.591s respectively.
asciilifeform: the most irritating numbskullism from this 1 imho is where he plugged his bot into human nick -- srsly, never noticed that no one else did this ? or, what, 'i'ma preshus snowflake and why shouldn't i'
asciilifeform: ( iirc claimed to have been tuned in silently for year+, even )
asciilifeform: enemy reaction to btc seems to resemble e.g. meat tryin' to digest an old bullet. 'maybe can't digest this but how about we coat it in immunoslime and maybe can pretend it aint there'
asciilifeform: 'yak' is the primary moving part in the 'taking' mechanism, neh. ( or at least the antibody marker... )
asciilifeform: afaik it's the original item of this type aha
asciilifeform: http://btcbase.org/log/2019-06-06#1917476 << oddly enuff asciilifeform always hated the search bar, and never used, it (or was it only in mine?) had irritating bug where your search string stick around on the display after use☝︎
asciilifeform: timing tests gotta use a snapshotted FG bin always, otherwise the variation in fg piss rate obscures the item being measured.
asciilifeform: ( i.e. cat tape.peh | ./bin/peh 32768 32 100000 0 dulap_1m.bin , in this scheme all '?' output is taken from the supplied file )
asciilifeform: this will be to determine what is the diminishing-return point (gcd with 1, 2, ... , 17 pieces , for prefilter, and 1st 1MB from fg sample bin on nosuchlabs.com, for deterministic test; see what diff makes to the time needed to gen N primes )
asciilifeform: while on subj : http://p.bvulpes.com/pastes/Mg63b/?raw=true << pieces of latest (not yet complete) experiment. when fed to a 32768bit peh run, produces the 32768-bit primorial (i.e. shows that the 17 2048bit cuts actually multiply to it )☟︎
asciilifeform: aand for that matter, if you were to put in a random turd instead of the primorial, you will still get valid primes, will simply take longer (i.e. the prefilter won't work)
asciilifeform: ( re upstack, note that primegen.peh can be written without any long constants, i.e. generate primorial at warmup. this simply costs a coupla extra min. )
asciilifeform: i intend to make it reasonably painless to learn, but will i suspect always require some sweat. rather like e.g. v.
asciilifeform: ftr anyone who thinks 'i'ma use other people's tapes blindly' is gearing up for world of pain. peh is less an object like gpg and moar like surface-to-air rocket, reqs some basic grasp of what yer doing
asciilifeform: http://btcbase.org/log/2019-06-06#1917441 << this is moar or less asciilifeform's exact position. mircea_popescu i rec. to glance at ch18, it has good picture of this (i.e. reader is not offered 'use this tape', is offered 'here's how to bake a primegen' ; instead of 'here's magic constant', given 'here's how it is made' etc )☝︎
asciilifeform: mircea_popescu: will let BingoBoingo answr when he wakes up, i dun think i have the latest #s
asciilifeform: ( can laugh at this hypothesis if you like, but i can't think of how else to explain the still-palpable amt of minerism that happens ~outside~ of cn and other cheap-mains locales )☟︎
asciilifeform: i.e. where folx will pay considerably moar than the typical 'market worth' in fiatola, simply to avoid moving fiatola into any box overtly 'bitcoin-labeled'
asciilifeform: fwiw i suspect that even today, a very similar logic drives some substantial % of minerism
asciilifeform: ... in 2010-11, i mined just short of single coin , on junkyard-salvaged fpgas ! until then, had none at all
asciilifeform: Mocky: if i had bottle of 'become mircea_popescu' elixir, i'd share with you a glass. but as it is e.g. asciilifeform was entirely unable to get hold of bitcoin at all for many yrs, when could not find anyone who had it + would not deal with goxisms☟︎
asciilifeform: ( typically they fold up and go straight-ebay when keeper wakes up to the fact that he pays rent for shop and sits in chair for mere 10% of his take... )☟︎
asciilifeform has been to various 2ndhand irons shops (industrial, entirely disinterested in konsoomer rubbish) over past 20y, and erry single one , where talked to proprietor, 'i have this shop, but 90% of revenue is ebay'☟︎
asciilifeform: sorta like how 'ebay' ~killed the secondhand-comps-for-cash market
asciilifeform: Mocky: also 'platform' to help folx atrophy their doing-without-inca-platform-ty muscles.
asciilifeform: BingoBoingo: imho really oughta be seen as same sort of item as 'uber' , 'airbnb', other 'inca will insert himself as middleman' schemes
asciilifeform: arguably was never anyffin other than a 'decentralization-flavoured' honeypot.
asciilifeform: ( obv. this method is NOT fastest means to compute product of 1st n primes; it is not exactly secret what the first coupla mil primes are, and one could compute product in <1s , if taking what they are on faith. but illustrated method does not rely on 'magic #s' )☟︎
asciilifeform: meanwhile, in upstack lulz, a 32768-bit primorial ! eats 63min 49s on tester, and holds 1st 2553 primes.
asciilifeform: meanwhile, in other lulz, putin published uncensored copy of molotov-ribbentrop pact .
asciilifeform suspects 'diminishing return' is 'logarithmic' here. but not proved yet.
asciilifeform: ... so in principle if you wanted to litmus against the 1st 1386 primes, in a 2048bit primegen , you will need 8 gcd litmusen, each with a separate 2048b constant.
asciilifeform: ...elementarily, the procedure where gcd(N, P), where P is a primorial, can be turned into two: gcd(N, P1) and gcd(N, P2) , where P1 and P2 ea. have half the bitness of the larger P ( recall effect of multiplication on bitness . )
asciilifeform: for thread-completeness : 16384bit fz : 1st 1386 primes. (took ~10min to compute , on ' asciilifeform's standard tester iron ', using tape shown in ch18 )
asciilifeform: in the didactic tape of ch18, i used strictly a single primorial litmus, with strictly 1st n primes that sit down in given width. but this is solely for simplicity of demo.
asciilifeform: q is , to what extent makes sense to do this.
asciilifeform: these can never be abolished (or who would need m-r then...) but can be reduced, to some extent, as one particular shot of gcd is substantially cheaper than a single shot of m-r☝︎
asciilifeform: ... unsurprisingly, vast majority of cycles of ch18 prime gen tape are spent on 'first shots' of m-r which return 'is composite'.
asciilifeform: for comparison : a 4096bit fz can hold product of 1st 418 primes ; a 8192bit fz -- the 1st 758 primes.
asciilifeform: mircea_popescu, diana_coman , other maffs folx ^☟︎
asciilifeform: meanwhile, in the torture room : 2048-bit primorial is product of 1st 233 primes, i.e. 2,3,5,...,1471 . not so many. theoretically it may be worth the sweat to include a multi-part primorial (i.e. ~set~ of gcd litmusen, each within the permitted width) . this gives correspondingly faster pre-mr sieving, at expense of coupla extra ln. of peh, per. at what point diminishing retu
asciilifeform: cuz yer 'amd f' spends 95% of its cycles waiting for ram.☟︎
asciilifeform: the other 'seekrit of the dark arts' is that on recent (~decade) of x86 irons, the gap b/w cpu an' ram clock is such that if you have sim that sits down in l0 cache, the speed penalty may well be effectively 0
asciilifeform: ( even aside from the moar obv. examples of what can be jettisoned, e.g. vga )
asciilifeform: notion here being, not all of the knobs on trad. physical box, are useful in a sim
asciilifeform: why exactly shouldn't yer session restart if you unhandled-div0.
asciilifeform: potentially. wouldn't want to live in a sim like this round the clock. but also had thought: so long as hypothetical controller for these logs the eggogs, it could easily be Right Thing for e.g. shared hoster.
asciilifeform: not really. like any other sim, it has reg dump
asciilifeform: largely unrelatedly, but really oughta put in l0gz before it vanishes into the sands : the sim-mips find from other day, i read it when went to bed, and author had vehehery interesting 'muntz' with which he made it short as it was: he dun handle any faults other than page fault. e.g. garbage instruction, div0, etc. simply terminate the sim run.☝︎
asciilifeform: i read the piece, but not seen film yet
asciilifeform: 'Understand that all is transient, the software that you write will probably not last your lifetime, and that it is philosophically impossible to write perfect software. It will always eventually begin to die, once the environment shifts enough. '
asciilifeform: going by the linked item, fella also ate just enuff logs to feed the cockroaches in his head, but not enuff to kill any, e.g.:
