Show Idle (>14 d.) Chans


← 2022-11-22 | 2022-11-24 →
billymg[busybot]: has anyone here used a whole house surge protector before? like this one: https://www.homedepot.com/p/Eaton-Whole-House-Surge-Protector-CHSPT2ULTRA-1/204761136
asciilifeform: billymg: do nontrivial surges actually happen where you live?
billymg[asciilifeform]: asciilifeform: all my electronics are plugged into either UPS or surge protector strips, and a few fridges and washing machines each have their own one of these
billymg[asciilifeform]: before moving here never heard of surge protecting large appliances. learned about it when a new fridge i bought had its motherboard fried a month later. when the store sent a technician to replace it under warranty they recommended the appliance surge protectors
billymg[asciilifeform]: asciilifeform: don't know for sure about surges but regularly notice e.g. lights dimming slightly, so the power definitely isn't clean and consistent
asciilifeform simply has a rack-mounted 'double convert' ups + expanders, to which ~errything of any value is connected
asciilifeform: has built-in MOVs for surge (tho would have to desolder if they were to blow. to date never happened.)
asciilifeform: also reports min/max voltages etc.
asciilifeform: ( subj fwiw. if you have moar than 1.8kW of load, would need fatter one.)
asciilifeform: with expander (and reasonably-fresh cells), ~40min of time at full load, moar than enuff time to change oil in generator.
billymg[asciilifeform]: asciilifeform: neat. for fridges i don't worry about 'uptime', just not frying them
billymg[asciilifeform]: since everything these days seems come with a motherboard
asciilifeform still has 'analogue' fridge, lol
asciilifeform: phf: moar re new crapples : on 'm2' fanless lappy, built ffa, and this benchmark ran in 1min 41s.
bitbot[busybot]: Logged on 2022-11-18 12:32:30 phf[awt]: meanwhile you have x86 market working with a very modular design, that necessarily creates overhead. so the whole "m1 totally slaughters legacy intel!!1" narrative is essentially a bunch of idiots going "bauaah duuuh standard bad, ipad render new york times fast good!!"
asciilifeform: phf: for comparison: 2m 49s on 'threadripper' 3960X.
asciilifeform: and, as given in orig. article, on 'opteron' 6328 : 7m 24.751s.
asciilifeform: phf: evidently the optimizations ~not~ 'crappleware-only'. ( and this was with unadorned gnat, w/out any crapple proprietary arm compilers. )
asciilifeform: pity they dun sell 'm2' in a fat rackable box, where could scrape off the crappleware and use as generic comp.
asciilifeform: currently appears to be the fastest (single-core, at any rate) cpu in asciilifeform's torture chamber.
asciilifeform was quite skeptical, and convinced 'benchmarks -- lies' etc. but troo phakt.
asciilifeform: notbad for a box that pulls ~20 watt (less when lcd off..)
asciilifeform: and, note, this was w/out asm inlines.
dulapbot: (asciilifeform) 2022-06-24 asciilifeform: for 'magic instrs' yer stuck with 'intrinsics' (of which gnat gives next to nuffin) and then of course asmism
asciilifeform expected the crapple to be warm enuff 'to fry egg' , at least, after this -- but nope.
asciilifeform ftr not found any anti-constanttime booby on 'm2'.
asciilifeform: the adhoc heathen gnat asciilifeform used on that box. (gnat in 'macports' appears to be 100% broken.)
asciilifeform to be fair not tried erry possible arm64; previously, only rk3328 (which goes at ~half the speed of opteron 6328, which still aintbad considering that it runs under <5W)
asciilifeform: x86 suxx, noose at 11.
asciilifeform: ftr, how to run: 1) throw out the flags (e.g. '-static') that dunwork on crapple gnat; 2) in limits.ads, up the max tape from 1MB to 12MB 3) up the stack to max crapple permits: ulimit -s 65520 3) :
asciilifeform: time cat tapes/10k_shots_1024bit_ffa_slid_rnd.tape | ./ffacalc/bin/peh 1024 32 10660000 0
asciilifeform: ( steps 2 and 3 needed on x86 also )
asciilifeform: ( why? cuz ch14 was before tape had to live in memory whole )
asciilifeform thinking about it, the above aint a 'universal' benchmark at all, as the calculation very much fits in cache (on all 3 boxes.) but still interesting.
asciilifeform: (i.e. fact that crapple's thing has ~1cm memory bus, not relevant here)
asciilifeform not has the arm 'neon' phf mentioned ; may be interesting to try on it
← 2022-11-22 | 2022-11-24 →