Show Idle (>14 d.) Chans


← 2020-08-21 | 2020-08-23 →
00:54 gregorynyssa http://logs.nosuchlabs.com/log/asciilifeform/2020-08-20#1020010 << what you are looking for is arguably not an FPGA, but a bag of LUTs. the interface between the LUTs and one's work-computer is something which he could implement by himself using TTL and PCB.
00:54 snsabot Logged on 2020-08-20 19:30:21 asciilifeform: there is presently no 100%-documented homogeneous fpga on market. and, for fundamental reasons, aint likely to be, 100% of fpga vendors live from the 'rent out the right to use our built-in NIC' and similar scamola
00:59 gregorynyssa problem is, no one sells densely etched LUTs as a generic commodity. everyone (Xilinx, Intel...) sells "solutions," or worse, "experiences," but I still see two narrow avenues out of this mess:
01:04 gregorynyssa (1) some universities have small VLSI facilities. see if you can get a deal to use those. (2) start your own coarse-grained fabrication-facility. 100-500 nm is fine for the next ten years of Lisp Machine revival. I don't think < 100 nm is necessary. presumably the costs of building such a facility have decreased compared to the 1980s.
01:06 gregorynyssa basically, the Armadillo Aerospace of fabrication.
01:11 gregorynyssa nuclear engineering students today know how to rebuild 1940s atomic bombs at a fraction of what it had cost that era. someone should likewise re-trace the machinery of 1985 from first principles.
~ 11 hours 28 minutes ~
12:39 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020093 << a fpga was originally ('80s) precisely 'bag of LUTs.' the 'let's include 9000 undocumented periphs and license out the closed turds to drive'em' biz model came in '90s.
12:39 snsabot Logged on 2020-08-22 00:54:11 gregorynyssa: http://logs.nosuchlabs.com/log/asciilifeform/2020-08-20#1020010 << what you are looking for is arguably not an FPGA, but a bag of LUTs. the interface between the LUTs and one's work-computer is something which he could implement by himself using TTL and PCB.
12:40 asciilifeform re: 'with ttl and pcb' -- ever seen the machines that were built this way ? y'know, the 'five refrigerator's worth of comp , + five actual refrigerators for cooling, tops out at 1MHz' items.
12:42 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020096 << re (1) -- if you aint a credentialed academic, or multimillion-$ donor, the conversation ends up ~very~ short.
12:42 snsabot Logged on 2020-08-22 01:04:21 gregorynyssa: (1) some universities have small VLSI facilities. see if you can get a deal to use those. (2) start your own coarse-grained fabrication-facility. 100-500 nm is fine for the next ten years of Lisp Machine revival. I don't think < 100 nm is necessary. presumably the costs of building such a facility have decreased compared to the 1980s.
12:43 asciilifeform closest i've come so far is some chinese uni who actually wrote to asciilifeform and offered use of their fab. but unsurprisingly wanted ~50k usd ~per run~ in compensation. folx dun work for phree, not even in china.
12:44 asciilifeform re (2) , 'start your own coarse-grained fabrication-facility' -- right after i open own spaceport, lol
12:44 asciilifeform 'presumably the costs of building such a facility have decreased compared to the 1980s' -- ~increased~.
12:44 asciilifeform substantially increased.
12:45 asciilifeform because of monopolization. at one time there were thousand+ semiconductor fabs on planet3. then, hundreds. now, perhaps two dozen.
12:46 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020098 << the cost of getting U outta the ground, and the qty of it needed, haven't changed ~at all.
12:46 snsabot Logged on 2020-08-22 01:11:29 gregorynyssa: nuclear engineering students today know how to rebuild 1940s atomic bombs at a fraction of what it had cost that era. someone should likewise re-trace the machinery of 1985 from first principles.
12:48 asciilifeform !w poll
12:48 watchglass Polling 12 nodes...
12:48 watchglass 205.134.172.26:8333 : Alive: (0.084s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 205.134.172.27:8333 : Alive: (0.089s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867 (Operator: asciilifeform)
12:48 watchglass 108.31.170.3:8333 : (pool-108-31-170-3.washdc.fios.verizon.net) Alive: (0.108s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867 (Operator: asciilifeform)
12:48 watchglass 205.134.172.4:8333 : (172-4.core.ai.net) Alive: (0.084s) V=70001 (/therealbitcoin.org:0.7.0.1/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 205.134.172.6:8333 : (172-6.core.ai.net) Alive: (0.143s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 208.94.240.42:8333 : Alive: (0.115s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 143.202.160.10:8333 : Alive: (0.233s) V=70001 (/therealbitcoin.org:0.7.0.1/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 192.151.158.26:8333 : Alive: (0.237s) V=70001 (/therealbitcoin.org:0.7.0.1/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 213.109.238.156:8333 : Alive: (0.333s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 188.121.168.69:8333 : (rev-188-121-168-69.radiolan.sk) Alive: (0.373s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:48 watchglass 103.36.92.112:8333 : (terebe.ns01.net) Alive: (0.587s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Blocks=644867
12:49 watchglass 176.9.59.199:8333 : Busy? (No answer in 20 sec.) (Operator: jurov)
~ 30 minutes ~
13:19 asciilifeform ftr also there can be no question of a 500nm fpga holding anyffin like a nontrivial cpu. ( for comparison, e.g. ice40 -- which is ~barely~ cpu-capable -- is a 40nm product. )
~ 43 minutes ~
14:03 gregorynyssa http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020099 << makes sense. I suspected as much, and I keep coming back to the 1990s as the decade when the development of computing went terribly wrong.
14:03 snsabot Logged on 2020-08-22 12:39:34 asciilifeform: http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020093 << a fpga was originally ('80s) precisely 'bag of LUTs.' the 'let's include 9000 undocumented periphs and license out the closed turds to drive'em' biz model came in '90s.
14:05 gregorynyssa http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020101 << yep. MIT's CAR and CADR, for instance. but I meant implementing your own peripherals for the FPGA using TTL, not an entire CPU. cf. the late 1980s CGA or EGA graphics adapters with the little eight-bit micro-controllers.
14:05 snsabot Logged on 2020-08-22 12:40:51 asciilifeform: re: 'with ttl and pcb' -- ever seen the machines that were built this way ? y'know, the 'five refrigerator's worth of comp , + five actual refrigerators for cooling, tops out at 1MHz' items.
14:12 gregorynyssa http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020125 << I guess the idea is decisively refuted, then..
14:12 snsabot Logged on 2020-08-22 13:19:48 asciilifeform: ftr also there can be no question of a 500nm fpga holding anyffin like a nontrivial cpu. ( for comparison, e.g. ice40 -- which is ~barely~ cpu-capable -- is a 40nm product. )
14:14 gregorynyssa asciilifeform: suppose we don't use any FPGA. suppose you directly owned a fabrication-facility and could perform your own runs on whim. how low does the process-width have to be, for it to be worth your time?
14:14 gregorynyssa (ie. for your project to succeed?)
14:15 gregorynyssa IIRC the 80386 had a process-width of 1,000 nm.
~ 27 minutes ~
14:42 asciilifeform gregorynyssa: for that matter, bolix's 'ivory' was a 2uM item.
14:44 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020128 << there's a kind of 'vicious cycle' there -- anyffin above 30-40MHz or so demands short, straight paths, and fast (ideally vlsi) logic on other end.
14:44 snsabot Logged on 2020-08-22 14:05:24 gregorynyssa: http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020101 << yep. MIT's CAR and CADR, for instance. but I meant implementing your own peripherals for the FPGA using TTL, not an entire CPU. cf. the late 1980s CGA or EGA graphics adapters with the little eight-bit micro-controllers.
14:46 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020132 << pretty big 'suppose'. 200nm is prolly the thickets practical (i.e. for achieving sumthing like early 2000s standards of throughput and i/o) . but i still can't resist to point out that this is rather like discussing 'slowest practical starship for alpha centauri colony'
14:46 snsabot Logged on 2020-08-22 14:14:33 gregorynyssa: asciilifeform: suppose we don't use any FPGA. suppose you directly owned a fabrication-facility and could perform your own runs on whim. how low does the process-width have to be, for it to be worth your time?
14:47 asciilifeform si fabs, whether in '70s, '80s, '90s, or today, were serious capital goods that had to be occupied w/ production of salable products 24/7 to make 'in the black'.
14:48 asciilifeform smbx for instance ultimately failed to pay back the investment in baking 'ivory'.
14:49 asciilifeform ( and they did, incidentally, have the good sense ~not~ to build own fab -- they had hp do it. and these were folx who blew own glass for crt's, as no off-shelf crt in 1980 would do 'supervga' res... )
14:50 trinque any attempt to fab own CPUs would have to step through many other, simpler devices on the way, or yes, pull in a huge investment.
14:51 trinque (with little chance of repayment)
14:51 asciilifeform trinque: aha. asciilifeform wanted to give a sketch ftr of just how 'megalomania' intrinsically in the idea of 'bake own fab'
14:51 asciilifeform it aint necessarily obvious to the innocent
14:53 asciilifeform it's an idea roughly on the scale of 'build own 10GW nuke power station'. and consider fact that e.g. su had working native designs for the latter -- but afaik never built ic fab w/ purely native tech, all contained smuggled-in heavy gear
14:54 asciilifeform even today, infamous 'elbrus' is fabbed in tw, at tmsc co.
14:57 asciilifeform ( incidentally, the ru design bureau that baked 'elbrus' did not even attempt to have the supporting periphs -- nic, vga, etc. fabbed . instead off-shelf. they figures, prolly correctly, that no amt of $ would suffice... )
14:57 asciilifeform * figured
14:59 asciilifeform 'holy grail' ultimate pill here would obviously be 'desktop fab'. but afaik currently is not even in realm of 'science fiction', but rather lsd fantasy, sadly.
14:59 snsabot (trilema) 2015-07-26 asciilifeform: folks like to talk about 'printed' this and that, but the actual hard data on the ground are considerably more optimistic for classical problems like 'lead to gold', 'elixir of immortality', and 'energy too cheap to meter' than for desktop chip-fab.
15:01 asciilifeform imho the standard methods for ic fab are a political dead end .
15:04 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-20#1020067 << btw i've taken the liberty of mirroring bvt's www .
15:04 snsabot Logged on 2020-08-20 20:55:45 trinque: http://logs.nosuchlabs.com/log/asciilifeform/2020-08-20#1019858 << meh, I'm done mourning these. waste abounds.
15:04 asciilifeform !q seen-anywhere bvt
15:04 snsabot bvt last seen in #ossasepia on 2020-05-25 05:15:47: if i liked that code, i would have just released the vpatch immediately with performance benchmark, so i agree with 'case against'. i will look into improving the error reporting code -- this should be possible (with slightly more code in v.sh), though the precise loop analysis would take approximately the same amount of ada code as went into that experimental vpatch.
15:06 asciilifeform if anyone has wish list of reasonably light-weight (bvt's is <4G for instance) sites to mirror, plox to write in.
15:07 shinohai !w probe 205.134.172.28
15:07 watchglass 205.134.172.28:8333 : Busy? (No answer in 20 sec.)
15:08 asciilifeform shinohai: that's whaack's noad. i'm presently satisfied that it's eating, does tend to answer 1 in ~20 queries , and monotonically climbs ( remember that trb is effectively single-threaded, it won't perform commands while verifying blox... )
15:09 shinohai ^ kk i was just testing was writing a blog post
15:09 asciilifeform a
15:17 asciilifeform e.g. from local box coupla min ago, got 205.134.172.28:8333 : Alive: (0.149s) V=99999 (/therealbitcoin.org:0.9.99.99/) Jumpers=0x1 (TRB-Compat.) Return Addr=0.0.0.0:8333 Blocks=448752
~ 1 hours 10 minutes ~
16:27 feedbot http://btc.info.gf/blog/the-caek-is-a-lie.html << btcinfo -- The caek is a lie
~ 1 hours 1 minutes ~
17:29 * trinque is over here grunting over the perl build for why it's shitting "pod" files into the root dir.
17:30 trinque after that it's about time to cut a genesis, I think.
17:31 trinque I'm curious what those present think in terms of a bootloader; I touched on this in the OS series before.
17:31 trinque lilo's small, but incurs its own assembler (bin86)
17:31 trinque extlinux looks like it *might* be small if it can be severed from the sprawling shitheap of syslinux (which at first pass demanded a python and other nonsense)
17:32 trinque grub eh, maybe an old one, but that's a tenuous maybe
17:33 trinque bootloader ftr is not needed for genesis. what I've got is plenty useful by itself as gcc-sturbator.
~ 2 hours 56 minutes ~
20:29 asciilifeform trinque: sometimes bootloader is not even needed at all -- e.g. on apu1 & other amd g-series boxen, can bake kernel straight into bios
20:29 asciilifeform so imho it is a secondary item, and asciilifeform for one would not object if it were genesis'd separately
20:30 asciilifeform the 1 problem with this is that linux kernel is 'pathologically lazy', in fact does not initialize irons properly, offloading good portion of this work to the loader
20:33 asciilifeform imho this oughta be fixed, at some pt, ~in the kernel~. really there is no reason why linux's bootloader needs to be any larger than my illustrative one.
20:33 snsabot (trilema) 2018-07-06 asciilifeform: ok here goes, ftr : http://loper-os.org/pub/x86-64-toystore-os.tar.gz >> sha512==e292a6d4296bc3cc63d2bc78bb7def807f7c4e9f8e630b292afec00b08c1fc2f8eeff5d074560804828ee7aee8ab5e43e698436c203c990d994863882e51446a
20:36 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020169 << imho a bootloader oughta fit in 512byte block, and is entirely acceptable to represent it as a script where e.g. xxd -r -p < << $stringofhexcoads . i.e. w/out an asmer dependency.
20:36 snsabot Logged on 2020-08-22 17:31:29 trinque: lilo's small, but incurs its own assembler (bin86)
20:38 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020165 << the sad fact is that just about erryone who programs computers eventually goes fishing, permanently. some folx sooner, others -- later.
20:38 snsabot Logged on 2020-08-22 16:27:43 feedbot: http://btc.info.gf/blog/the-caek-is-a-lie.html << btcinfo -- The caek is a lie
20:38 snsabot (trilema) 2014-11-04 asciilifeform: as a student, i was once told by a greybeard: 'you're young, but know that you have X lines of code in you. after that - log cabin.'
20:39 * asciilifeform doesn't see this fact as sumthing to be ashamed of. but would like to accomplish a few useful things before his filament burns out.
20:42 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020166 << which linuxisms have hard-dep on perl ? ( asciilifeform is vaguely aware that there were some, but can't recall immediately )
20:42 snsabot Logged on 2020-08-22 17:29:16 trinque: is over here grunting over the perl build for why it's shitting "pod" files into the root dir.
20:43 asciilifeform http://logs.nosuchlabs.com/log/asciilifeform/2020-08-22#1020171 << grub imho is simply junk and oughta vanish like bad dream. ( my irons afaik all at this point grub-free . )
20:43 snsabot Logged on 2020-08-22 17:32:35 trinque: grub eh, maybe an old one, but that's a tenuous maybe
20:44 trinque at least automake wants perl, and there were several others.
20:44 asciilifeform aah
20:44 trinque while automake is still used by gcc, gotta have perl, sadly.
20:44 trinque I don't intend to genesis something that "so long as you don't blow on it too hard, don't need x y z"
20:44 * asciilifeform doesn't disagree
20:45 trinque ftr gprbuild seems a good candidate for blowing it all away.
20:45 trinque (perl, m4, autoconf, automake, libtool)
20:46 trinque the parts I've contributed so far only want posix shell
20:46 asciilifeform trinque: gprbuild i can see replacing standard 'gnumake', but it doesn't (by design) have generalized text-munging, so prolly not m4 et al, unless somehow the 'need' for munging can be eliminated
20:46 trinque I'm pointing to eliminating the need, yes.
20:47 asciilifeform aha
20:47 asciilifeform observe that the folx who baked ada, apparently agreed w/ asciilifeform's take on 'ifdefism'
20:47 asciilifeform there's 0 support for it in the lang, or even in 'gprbuild'
20:47 trinque indeed, and god help me if I start caring about quirks of make on solaris, or wtf.
20:48 * asciilifeform can even see the pov of folx who want different build paths for $proggy on x64, arm, arm64, etc. but ada committee apparently did not.
20:50 asciilifeform 'portabilityism' is a traditional concern for unix folx, but evidently not for the people for whom 'there's 1 rad-hard 386, circa 1990, and guess what, this satellite WILL use it'
20:52 asciilifeform the interesting bit is that ada has features like hard-bitness constraints for structures, native support for endianisms (including multiple endianisms ~inside one data structure~) and similar support for troo portability afaik absent ~anywhere~ else .
20:53 asciilifeform but the way i understand it, the above were included simply from pov of 'a correctly-writen program MUST maximally specify its semantics', rather than from 'what if someone wants to build this for a vax suddenly'.
~ 33 minutes ~
21:27 feedbot http://mvdstandard.net/2020/08/ussa-proud-boys-make-first-appearance-in-portland-unrest/ << The Montevideo Standard -- USSA: "Proud Boys" Make First Appearance In Portland Unrest
~ 33 minutes ~
22:01 shinohai $vwap
22:01 BusyBot The 24-Hour VWAP for BTC is $ 11631.46 USD
22:10 trinque shinohai: too stable, snoar
22:10 shinohai lel
22:20 trinque asciilifeform: yep, this is the difference between design and accidental evolution.
~ 25 minutes ~
22:46 trinque shinohai: ftr, if you ever go fishing without coming back, I suspect you'll have the balls to say adios.
22:46 trinque I don't fault whatever someone chooses to do with his life. there's a shortage of people who can choose anything.
22:47 trinque hard to respect the silent fade-out, cowardly turn-off of the blog and irc with moist upper lip.
22:47 shinohai But of course. If I don't say adios, then you know i'm ded or ussa politburo has me.
22:48 trinque heh, we'd have to be way more interesting for any of that to happen.
22:49 * trinque still marvels, bitcoin didn't get uninvented, computers didn't stop their slip into oblivion.
← 2020-08-21 | 2020-08-23 →