UmbralRaptor changed the topic of #kspacademia to: https://gist.github.com/pdn4kd/164b9b85435d87afbec0c3a7e69d3e6d | Dogs are cats. Spiders are cat interferometers. | Космизм сегодня! | Document well, for tomorrow you may get mauled by a ネコバス. | <UmbralRaptor> … one of the other grad students just compared me to nomal O_o | <ferram4> I shall beat my problems to death with an engineer.
e_14159 has quit [Ping timeout: 200 seconds]
e_14159 has joined #kspacademia
<UmbralRaptor>
Best reason to use stainless steel: can use soap, don't need to worry about scratching it like teflon. https://xkcd.com/1905/
icefire has quit [Read error: Connection reset by peer]
* UmbralRaptor
vaguely wonders if half the channel will be playing Opus Magnum over the weekend.
<bofh>
That's the new Zachtronics, right?
<UmbralRaptor>
Yep
<bofh>
!u ???
<Qboid>
U+10903 PHOENICIAN LETTER DELT (?)
<Qboid>
U+10902 PHOENICIAN LETTER GAML (?)
<Qboid>
U+1090D PHOENICIAN LETTER NUN (?)
<bofh>
!u ???
<Qboid>
U+1202D CUNEIFORM SIGN AN (?)
<Qboid>
U+12055 CUNEIFORM SIGN DA (?)
<Qboid>
U+120F6 CUNEIFORM SIGN GAN (?)
<bofh>
I'm not sure that first one (the an) is needed, but let me confirm with the archaeological data
<UmbralRaptor>
?
<bofh>
oh I'm trying to figure out the Akkadian form of the Semitic fish-deity Dagon
<UmbralRaptor>
Ah
<bofh>
b/c I inexplicably appear to somehow be very good at nerdsniping myself into researching Mesopotamian archælogical facts tonight.
<bofh>
archæological*
<bofh>
typing/hard/shopping
<rqou>
huh, hexchat has pretty bizarre RTL text selection behavior
<rqou>
!u xxxxxעִבְרִיתxxxxx
<Qboid>
rqou: Too many characters! (Maximum: 10)
<rqou>
!u xxעִבְרִיתxx
<Qboid>
rqou: Too many characters! (Maximum: 10)
<rqou>
!u xעִבְרִיתx
<Qboid>
rqou: Too many characters! (Maximum: 10)
<rqou>
!u עִבְרִית
<Qboid>
U+05E2 HEBREW LETTER AYIN (ע)
<Qboid>
U+05B4 HEBREW POINT HIRIQ (◌ִ)
<Qboid>
U+05D1 HEBREW LETTER BET (ב)
<Qboid>
U+05B0 HEBREW POINT SHEVA (◌ְ)
<Qboid>
U+05E8 HEBREW LETTER RESH (ר)
<Qboid>
U+05B4 HEBREW POINT HIRIQ (◌ִ)
<Qboid>
U+05D9 HEBREW LETTER YOD (י)
<Qboid>
U+05EA HEBREW LETTER TAV (ת)
<rqou>
hmm, yeah
<rqou>
it's consistent
<rqou>
hexchat will always do an LTR selection, and then the rtl content will dance around inside the selection
<rqou>
hmm, it's actually even more broken than that
<kmath>
<FioraAeterna> if you can find the bug here without the debugger pointing you to it, you're a better programmer than me https://t.co/tIoWMJdeBp
<egg>
!wpn Fiora, whitequark, et al.
* Qboid
gives Fiora, whitequark, et al. a specific prepromorphism
<egg>
Fiora: I don't get why it ends up screwing with the stack though
APlayer has joined #kspacademia
<soundnfury>
egg: yeah, /me sees the second Score += it->second; should use it2
<soundnfury>
and it took me a good minute-and-a-half to spot it, but that's because C++ is unreadable to me
<egg>
it's unreadable to me too, and I write it both at work and in principia
<egg>
but then the preceding tweet suggests that this results in something caught by asan and I really do not see how
<egg>
(all hail asan though)
<soundnfury>
well it seems like there's a read off the end of a (vector?)
<soundnfury>
accessing unallocated memory sounds like the sort of thing you'd eggspect an Address SANitiser to catch
<egg>
oooh right, we might be looking at it even if it's at end
<egg>
I had noticed the discrepancy, not the UB implication
<egg>
(not a vector btw, some manner of associative container)
<soundnfury>
oh yes
<soundnfury>
but the .end() makes it sound like said container is just a std::vector of key-value pairs that get linearly searched, or some such horrid thing
<egg>
nono, maps have ends
<soundnfury>
a _sensible_ map-find would return NULL for not found :P
<egg>
even unordered maps
<egg>
I guess if one reinvented the STL now it might return an optional, but, C++
<egg>
anyway, I should poach an egg or two
<soundnfury>
!wpn egg
* Qboid
gives egg a RF hydrofluoroolefin which strongly resembles a vibrator
<soundnfury>
in case you run into any gamekeepers
<soundnfury>
(eggkeepers?)
<Fiora>
egg: somehow when run in lldb it crashed way later, and when i tried to debug, it failed
<Fiora>
it was ASAN that found the problem's source
<Fiora>
soundnfury: it's a DenseMap
<soundnfury>
seems pretty dense to me :P
<rqou>
C++ is terrible, news at 11 :P
<egg>
all languages are terrible
<egg>
!wpn Fiora
* Qboid
gives Fiora a sodium mine
<rqou>
um...
<egg>
!wpn rqou
* Qboid
gives rqou a smoked Hamiltonian
<rqou>
wtf lol
<rqou>
!wpn egg
* Qboid
gives egg a laevorotatory icosahedron
<rqou>
is there a way to list the words !wpn knows?
<egg>
hm
<egg>
!wpn -list
* Qboid
gives -list a commuting invariant
<egg>
nope
<egg>
!wpn -stats
<Qboid>
egg: Total weapons: 682. Total adjectives: 877. Total possible combinations: 1433118227750.
<egg>
rqou: ^ and then it's just the stamp collector problem :-p
<rqou>
huh, i always learned it as the coupon collector problem
<egg>
TIL in french it's vignettes
<rqou>
welcome to the land of cache invalidation, naming things, and off by one errors i guess :P
<rqou>
the two hard problems of CS
<egg>
eh on the naming things I'm not too bad with my catbus at work and, well, the Principia codebase here :D
<rqou>
btw, can anybody here draw me a venn diagram of OpenGL, OpenGL ES, DirectX {8|9|10|11|12}, WebGL, and Vulkan?
<egg>
maybe Fiora
<rqou>
why does it seem like graphics is the worst part of computers?
<egg>
though she might 3d print the diagram
<rqou>
anyways, i'm toying with the idea of using Rust+WebGL (via asmjs)
<rqou>
basically three different layers of experimental stacks :P
<kmath>
<FioraAeterna> basically programming is bad and i really want to get back to cosplay which is less bad
<soundnfury>
rqou: for the love of all that's optimised, WHY?
<rqou>
because *) Rust > C++ (i hate C++)
<soundnfury>
egg: |(sinplay, cosplay)| == 1
<soundnfury>
rqou: We all hate C++. That doesn't make Rust OK!
<rqou>
*) javascript+html5 is (for better or worse) the most portable "app framework" nowadays
<soundnfury>
stabbity
<e_14159>
egg: Wait, is that s/programming/math or s/cosplay/math?
<rqou>
and *) nobody made WebVK yet :P
<egg>
e_14159: cosplay
<rqou>
i actually like Rust
<rqou>
i hate web tech, but native gui tech sucks even harder
* egg
should try the fungus language someday
<rqou>
and i hate graphics, but what can you do? software renderers aren't quite fast enough
<soundnfury>
rqou: you can... write things that don't need graphics?
<rqou>
right, but sometimes i really want graphics
<rqou>
i've been meaning to learn Vulkan (for native, not the web stack) but last i looked there basically weren't any good tutorials
<soundnfury>
thou hast strayed from the path of Ritchian minimalism. For thy penance I prescribe two hours of /bin/ed.
<soundnfury>
;)
<rqou>
and i didn't care _that_ much, so i haven't invested the time to learn it the hard way
<rqou>
apparently nobody bothers to write tutorials geared towards somebody like me:
<rqou>
*) not a linear algebra tutorial *) not an operating systems tutorial *) not a computer architecture tutorial *) but have never touched anything at all related to graphics
<rqou>
because anybody who would want something like that is already capable of learning it the hard way :P
<egg>
rqou: I can't provide that, but I can provide a saner linalg tutorial than you tend to encounter :-p
<egg>
or maybe more maddening, rather than saner
<egg>
same thing really
<rqou>
i really want that for tensors
<rqou>
i can survive with my current linalg understanding
<rqou>
but it's 3am and i'm about to go to sleep
<soundnfury>
!wpn rqou
* Qboid
gives rqou a giant |raw⟩
<egg>
ah yes that's a good time for sleep
<egg>
!wpn rqou
* Qboid
gives rqou a plane Orcrist
<egg>
$%^@#$% I just finished my eggs and I forgot the bacon in the kitchen
<kmath>
<ManningKrull> Nothin', & I mean nothin', makes me madder than those animal skeleton decorations where they got bones where there… https://t.co/1ysi0rk8eX
<UmbralRaptor>
xShadowx: depends on the surrounding environment? Bits could travel a long ways in vacuum, antimatter ones not so much in an atmosphere.
<egg>
UmbralRaptor: you mean you couldn't distinguish the 50ish birb hieroglyphs if skeletonized?
<egg>
bofh: hm, at what daily dose (measured in espressi for want of something more accurate) are headaches eggspected in the absence of coffee?
<bofh>
in theory even one suffices, I think it's more an issue of how long and consistent the dosing is. but 2-3 are when it'd be apparent.
* egg
forgot to take his depakine yesterday evening, but even by those standards today's headache was particularly persistent; also forgot to make coffee at breakfast, hence the question
<egg>
bofh: yeah I pretty much *always* have ~1.5 in the morning, and ~a couple more during the day at work (less deterministic on week-ends)
<APlayer>
egg: Have you tried numerically approximating that value?
<egg>
!wpn bofh
* Qboid
gives bofh a Kahan yxala which strongly resembles a series
<egg>
!wpn Ellied
* Qboid
gives Ellied a zeta ␕
<bofh>
(so the other stimulants I take usually provide enough vasoconstriction to ward off the caffeine withdrawal headache for a few hours. you'd think this would be a good thing but instead it means that my first caffeine withdrawal symptom is "inexplicable listless sadness", which is sufficiently subtle and nonspecific enough it took me ages to realize it meant I need another damn coffee)
<kmath>
<daveaddey> And if one of the localized languages is German, sooner or later you’re going to have to switch to a narrower font. https://t.co/ddhNf7PM0D
<egg>
it... kinda feels reasonable, or at least consistent with what I've experienced toying with principia?
<bofh>
"Its emesal pronunciation was dimer." LOL
<egg>
which is good, it means principia is effective at instilling some manner of intuition of what those trajectories should look like :-p
<bofh>
Now I want to know what Sumerian glyph was pronounced "tetramer" :P
<APlayer>
How come does it bend downwards in the upper right corner?
<APlayer>
Is that a frame of reference thing I am missing?
<egg>
remember that we're in a rotating frame
<egg>
things that go away spiral like this
<APlayer>
So it is basically a spiral-unwind thing?
* Iskierka
is looking around for info on regularised least squares for machine learning, finds lecture slides that say it makes no sense for this purpose but works great for some reason
awang has joined #kspacademia
<egg>
or if you want to work directly in this frame, this appears to have done a loop around L2
<APlayer>
egg++
<APlayer>
This is amazing
<UmbralRaptor>
Iskierka: machine learning in a nutshell?
<bofh>
if I key in "???" into google, one of the results is, AND I QUOTE, "dagon fish god starbucks"
<bofh>
in the autocomplete that is
<Iskierka>
I think a number of other things probably were intended for the machine learning purpose in the first place, even if no-one really understands what it's doing
<Iskierka>
(amazingly, even the stupid idea of regLS against the test data labels instead of binary classification thislabel/notthislabel actually gave not-terrible results in recognising digits. 3 tended to be lower guesses, 6 middle, and 8 higher)
<bofh>
awang: nice!
<Iskierka>
it was absolutely wrong in its guesses, but it somehow actually managed to capture the trend in image recognition
<awang>
They also had slideshows of the images on the record
<awang>
And a copy of Murmurs of Earth
<awang>
And Newton's Systems of the World
<awang>
And a lady complaining about the lack of images of pyramids on the record
<Iskierka>
I feel like the great pyramids of egypt would have very little significance in simple images with no scale
<Iskierka>
Teotihuacan Pyramid of the Moon and similar central american/south east asian staggered pyramids would probably have more significance, as any aliens would likely have built in such configurations in their ancient times and would be able to recognise steps
<Iskierka>
so it would contextualise scale compared to the builders to be able to see a stepped path
<egg>
!wpn Iskierka
* Qboid
gives Iskierka a Александров walrus
<APlayer>
Qboid still has grammar troubles when dealing with Russian
<egg>
well vowel detection is kinda broken even in latin; as for lack of declension this is working as intended I think: in english you don't decline https://en.wikipedia.org/wiki/Alexandroff_extension (and I don't use transliterations because transliterations are bad)
<kmath>
<whitequark> "… I have no idea whether the guy is called Chebyshev, Tchebyshoff, Tschebyscheff or whatever" -- @eggleroy https://t.co/BzvVv8otZP
<UmbralRaptor>
egg: what do you use for the polynomials named after him? :D
<awang>
Iskierka: There were scale markers on the images
<APlayer>
I don't think that's presentable, though
<awang>
Iskierka: The lady also mentioned Mayan Pyramids, if that counts :P
<awang>
She was rather distressed over the entire thing
<Iskierka>
Then I do agree with her on that part. Though without actually knowing the images that are there I can't say if I'd be as concerned as her
<awang>
Some of them are surprisingly ordinary
<awang>
Woman raking leaves under a rather large tree
<awang>
Cars on Route 13
<Iskierka>
It does make sense to contextualise life on earth
<awang>
Woman buying grapes at the local grocery store
<awang>
At least those were ones she complained about
<awang>
Yeah, that's what some of the other visitors were saying
<Iskierka>
if we nuke ourselves and they come find us later, all they'd have is archaeology to base stuff on, so they'd be able to find stuff like pyramids
<Iskierka>
what they don't know is how our society works
<APlayer>
Well, we do have some idea of, say, the ancient Egypt or Maya societies
<Iskierka>
as much as pyramids would be good pictures to have, it is important to preserve what will not be preserved
<awang>
That's a good point
<awang>
I should probably go back downstairs and see if she's still there
<Iskierka>
We have some idea. It's a relatively recent revelation that the pyramids were likely not built by slave labour. And much of what we know is able to be determined due to the rosetta stone - can we be confident any similar artefact will exist?
<Iskierka>
how much are we even making that will preserve? Paper will decay quickly enough. Part IDs machined into the part are the only thing on the top of my head that archaeologists are likely to find
<APlayer>
I guess some sort of digital data on punched cards will be preserved in someone's nuke safe basement :P
<egg>
(also how is the search for additional kitens proceeding)
<Iskierka>
and how long will those punchcards survive?
<whitequark>
egg: pretty badly
<Iskierka>
even the rosetta stone is damaged
<egg>
whitequark: :-\
<egg>
whitequark: annoyed by порошок, or other issues?
<APlayer>
Well, don't know. We sent the golden records with the Voyager probes, I guess something similar is located on Earth in a few places
<Iskierka>
and will then have to be found
<awang>
Wasn't there a foundation of some sort dedicated to preserving knowledge for cases like this?
<awang>
Long Now or something like that?
<APlayer>
"Preserving knowledge for cases like this"
<APlayer>
We are talking about a total eradication of the human race
<whitequark>
egg: the former mostly i think
<APlayer>
There will be noone around to discover the knowledge, at least not for the time we may preserve it in any way
<APlayer>
Even if we badly wanted to
<Iskierka>
Digital data will decay over time, and even if found in time, they would have to determine our standards without damaging the samples. Paper and similar will decay. Only information marked in something solid with permanence, like machined parts or cut stone, can have any real confidence of being preserved long enough
<APlayer>
I don't think a diamond plate as used in Stephen Baxter's Voyage would last enough for either an extra-terrestrial race discovering it or any newly-evolved life on earth
<Iskierka>
hell, your best best for saving information for later explorers might be something like a simple description of the QR code standard, then machined QR codes describing either data interfaces, or data directly
<egg>
whitequark: and still no new kitten?
<egg>
blarg
<APlayer>
Iskierka: And how the hell would you describe the QR standard to some future explorers?
<Iskierka>
It's a representation of binary, binary is universal. Have a blank QR code for example, point out marking orientation with "this way up" symbol that should be easy enough to parse, then directions for parse-path of QR code
<whitequark>
that is actually pretty hard
<whitequark>
i mean have you seen the algorithm
<egg>
whitequark: the algorithm to find kittens?
<egg>
ah you mean QR?
<Iskierka>
well okay, QR code itself may not be good as it's not pure binary representation, but a similar design could be used. Thought is just it's a compact method of storing data we could use to describe how to interface with HDDs, if you want to use hermetically sealed ones for long-term mass data preservation
<Iskierka>
Basically, just "if we want to preserve as much as possible, how do we bootstrap discoverers to the densest long-term method we can manage"
<APlayer>
I'd go for a three nested gas-proof sealed metal chambers with a diamond engraving inside
<Iskierka>
I don't see how that would be any better than just machined stainless steel
<Iskierka>
even if you can do better precision on the diamond possibly, you're sealing it away with a lot of overhead
<APlayer>
Well, unless something catastrophically impacted it, it would slowly decay through three layers before even touching the diamond
<Iskierka>
which would be very expensive compared to just a few milled blocks of steel put in some more "meh, it's closed"-quality bunkers around the world
<APlayer>
Anyway, nice talk, and I'd resume it at a later point, but I've got to go for now
<APlayer>
Iskierka: I am talking about erosion proof for millions of years
<Iskierka>
I feel like it might be being pure carbon
<APlayer>
We seal it in a noble gas atmosphere
<APlayer>
:P
<awang>
And wouldn't something like gold be better for chemical inertness?
<Iskierka>
and over that timescale subduction is probably your biggest threat, so the best counter would simply be lots of samples. Meaning you want low-cost durability
<awang>
I thought thermodynamically unstable means it'll spontaneously decompose
<awang>
Regardless of chemical reactions
<APlayer>
Well, let's plate it with platinum
<Iskierka>
gold's softness would be a problem for erosion
<awang>
I don't think platinum binds to diamond...
<awang>
Gold-plated?
<awang>
But yeah, softness would be an issue
<APlayer>
Okay, I've got to go for real this time
<APlayer>
See you!
<awang>
\o
<Iskierka>
although if cost were a non-issue to the point of considering large amounts of gold or diamond, fuckin' inconel man. It'll destroy the thing trying to break it first
<APlayer>
;tell me the outcome of your theories
<awang>
I forgot that inconel was a thing
<Iskierka>
oxidation, reduction, stress, heat, cold, invonel don't give a duuuuuuck what you try do to it
<Iskierka>
hell its normal use is resisting the whole first four all at once
* Iskierka
presumes Mehrunes Dagon, Daedric Prince of Destruction, also representing revolution, energy, change, and ambition
APlayer has quit [Ping timeout: 383 seconds]
ferram4 has joined #kspacademia
awang has quit [Ping timeout: 198 seconds]
awang has joined #kspacademia
<egg>
!wpn bofh
* Qboid
gives bofh a multipole slow reflector-like LED
<egg>
!wpn whitequark
* Qboid
gives whitequark a monopolar hypothesis
<egg>
!wpn котя
* Qboid
gives котя a nitrogen principle
<Iskierka>
... I managed to misread that twice, first as particle, then as participle
<bofh>
Fun discoveries: I used to write past assignments using 0.3mm 3H mechanical pencil lead. Yes, it exists.
<egg>
bofh: yeah, it has the advantage of being less smear-prone though
<bofh>
egg: oh I agree, I can't use pencil lead softer than 2H. Just... 0.3mm? I basically used that as an excuse to treat some of my assignments as optimization problems for how small I can write and still have it be legible.
<egg>
:D
<bofh>
The answer to how small is fitting two lines of text in a single line of standard 8.5x11" lined paper pretty consistently in some cases, at least until the TAs told me to cut that shit out.
<egg>
(we being phl, I played little part in that)
<awang>
Gah
<egg>
Fiora: they've been getting better tbh
<awang>
I tried to right-click, not copy-paste
<Fiora>
well, i mean, yes, for one they threw out the compiler and replaced it with llvm
<awang>
idk how that happened
<Fiora>
which is a good start
<awang>
Wait what
<egg>
Fiora: well that's still optional
<awang>
When did that happen
<Iskierka>
writing C++ sounds like an adventure
<Fiora>
a year or two ago
<egg>
clang/c2 is an option but not the default
<Fiora>
Frontend is still MS
<awang>
wut
<awang>
That sounds very not-Microsoft
<Fiora>
awang: dark secret, custom compiler backends are dying
<egg>
oh you mean the *back* end?
<Fiora>
because maintaining 15 year old legacy codebases makes all your employees leave
<Fiora>
er, midends
<Fiora>
backends are dying too but midends are dying faster
<egg>
so c2 is llvm-based you mean?
<egg>
I'm confused
<Fiora>
yes....
<egg>
so they have their own frontend, and also you can use clang as an option
<egg>
and they have their own posterior C2? but that's llvm-based?
<bofh>
(Hilarious thing about Intel: icc and ifort have *totally different* backends, apparently icc's is deprecated. ifort's is not).
<bofh>
(like, totally different origins even. the latter derives from decfort90, the former was always Intel's thing).
<awang>
bofh: I never would have guessed that o_O
<awang>
That sounds like a pain to maintain
<awang>
One compiler is a lot of work as is
<Fiora>
huh. looks like they're killing off clang/C2 entirely
<egg>
oh?
<awang>
egg: VS 2017 for Principia soon then?
<egg>
awang: pending release I guess
<Fiora>
as in it's deprecated
<bofh>
awang: I only discovered it b/c awhile ago I setup decfort90 on an Alpha and discovered it had so many of the same quirks as ifort, ranging from identical warning strings to similar high-level codegen quirks.
<egg>
because that incorrect compilation was killing everything with our geometry
<Fiora>
and now it's just msvc or straight clang? not sure. conflicting info
<kmath>
<StephanTLavavej> The Clang/C2 experiment for C++ is ending. Our STL is now tested against Clang/LLVM 4.0. (VS 2017′s second toolset update might need 5.0.)
<awang>
bofh: Why were you setting up decfort90 on an Alpha in the first place?
<egg>
Fiora: huh; (insert joke about the tandem bicycle with the messengers carrying orders and counterorders)
<awang>
egg: What joke is that?
<bofh>
awang: compiling fortran code that linked against an alpha-only binary quantum chem lib (and I had an alpha machine b/c reasons).
<bofh>
tl;dr academia
<egg>
awang: not sure where it's from (I probably know it from my grandfather?) but well, description says it all :-p
<egg>
bofh: I've heard good things about the Alpha
<awang>
bofh: ...scary
<awang>
How old is that Alpha?
<awang>
egg: I still don't get it :(
<bofh>
egg: good points: it was fast, it was novel, it really was ahead of its time. bad points: its cache model violates causality, 100% ECL logic meant it doubles as a space heater.
<Fiora>
egg: yeah, i have no idea what's up with all this, it's weird
<Fiora>
e.g. i heard rumors ICC was switching to llvm
<Fiora>
at least as a mid-end
<Fiora>
also "mid-end" is the dumbest term
<Fiora>
bofh: ECL logic?
<egg>
awang: two messengers (in the military) riding a tandem bicycle, the one in front is carrying the orders, the one in the back the counter-orders
<bofh>
Fiora: emitter-coupled logic as opposed to CMOS logic.
<egg>
Fiora: yeah I was going to say, what does "end" even mean Ꙩ_ꙩ
<awang>
bofh: Could you expand more on the Alpha cache model? This sounds amusing
<awang>
egg: ....Oh
<bofh>
Fiora: basically your logic gates are made of overdriven single-ended BJT differential amplifiers
<Fiora>
wtf
<Fiora>
did they manage to make something more power-hungry than domino logic
<egg>
bofh: causality is overrated though
<Fiora>
"the input/output voltages have a small swing (0.8 V)" oh gods
<bofh>
Fiora: so it's sort of the inverse of domino logic I think. iirc, domino logic has rail-to-rail swing?
<Fiora>
is this one of those "every transistor is a ULVT" thing
<bofh>
in ECL the swing is tiny
<bofh>
and the logic is current-driven
<Fiora>
static leakage ahoy?
<bofh>
what leakage? the current draw is essentially constant.
<Fiora>
That *is* leakage, isn't it?
<Fiora>
"static leakage" == power spent when not flipping state
<Fiora>
so i presume this means zero dynamic power, *all* static power
<bofh>
pretty much.
<Fiora>
so, funny related thing
<Fiora>
which i can now say, i suppose
<whitequark>
ECL is super hot
<Fiora>
so the apple GPU silicon team wasn't always a GPU team
<Fiora>
their original specialty, prior to being bought out, repurposed, and expanded, was actually physical design
<whitequark>
like you can't touch ECL chip casings without burning yourself
<whitequark>
typically
<bofh>
(I actually measured power draw on an AlphaServer CS20 once. 330W idle. 360W at 100% CPU usage on both cores).
<Fiora>
their physical design specialty was domino logic
<bofh>
s/cores/chips/
<Qboid>
bofh meant to say: (I actually measured power draw on an AlphaServer CS20 once. 330W idle. 360W at 100% CPU usage on both chips).
<Fiora>
now for sure there's probably not much of the original team left at this point, though i don't actually know the numbers
<Fiora>
i know there's some ex-AMD Cat team there (okay, to be fair, they're everywhere now, except AMD)
<bofh>
whitequark: I've actually made tea on an Alpha's heatsink. turned off half the fans blowing across it then made some pretty okay-tasting green tea in a glass sitting on top of it.
<bofh>
I stress this was with half the fans still active.
<bofh>
Fiora: heh. what happened at AMD that caused them all to leave, anyhow?
<bofh>
assuming that's not NDA'd to hell
<whitequark>
lol
<Fiora>
bofh: they were laid off
<Fiora>
lol
<Fiora>
AMD laid off the team after Jaguar
<bofh>
Huh, that's interesting.
<Fiora>
good job folks, lay off the team that's made your CPU group more profit in 10 years than any other chip
<Fiora>
they took about half into Bulldozer and killed the project
<Fiora>
and the rest were laid off
<Fiora>
now they work at Every Other CPU Company
<egg>
bofh: okay I'm pretty sure being able to make tea on it should be considered a *feature* of the Alpha :-p
<bofh>
Jaguar struck me as a fairly good low-power/portable µarch
<bofh>
(I have one of those in what I affectionately refer to as my craptop (a shoddy netbook), it has weirdly okay perf that sometimes makes no damn sense, like the critical instructions often are very different from the ones I expect (my expectation is that which you'd get writing SIMD asm primarily for HSW))
<Fiora>
Jaguar isn't bad at all
<Fiora>
SIMD-wise it's actually the most boring, sensible arch of any x86 ever made
<Fiora>
that is, almost everything has 1/0.5 latency/inv throughput
<Fiora>
except for things like multiplies and so on, which are mostly terribly predictable
<Fiora>
it has a few oddities, but *so much* less than intel
<Fiora>
the end result is that for CPU code, it's fantastically easy to get 2 instructions per cycle in optimized code because of this lack of bottlenecks
<Fiora>
the downside of this is that you can get, as most, 2, because it's overprovisioned on nigh everything except for frontend
<Fiora>
this makes it Boring since everything is just 2, only 2, no less, no more
<kmath>
YouTube - Baby jaguars caught on camera - Natural World: Brazil's Super Cats: Preview - BBC Two
<bofh>
Fiora: oh, another fun thing about ECL is it let you do crazy things like *off-chip L2 clocked at the same freq as the core* (and apparently in the EV3/EV4, *OFF-CHIP L1*).
<bofh>
let me get a citation for the latter since it's absolutely insane
<bofh>
"The 21064/21064A is a high performance chip that has some stringent thermal characteristics which need to be considered when evaluating a method of cooling the device." wow you don't fucking say
<egg>
bofh: well it's easy, just continuously run tea over the heatsink
<bofh>
okay apparently internally the 21264 was mostly CMOS, but connected out via ECL to the memory controller and all the cache, so it's more correct to say the ALU and register file is CMOS and the memory controller and cache controller is ECL.
<bofh>
which... doesn't mean much since load/stores and routing eat most of your power draw anyway.
<egg>
speaking of tea I feel like drinking some
* egg
brb
<egg>
bofh: and see, this machine an Alpha so I need to walk all the way to the kitchen to boil some water :-p
<bofh>
(actually my favourite thing is that the manual for the AlphaServer CS20 actually lists a heat dissipation figure of 939BTU/hr in the *first page of specs*).
<egg|tea|egg>
;rpn 1 1054 * 3600 /
<kmath>
egg|tea|egg: 0.2927777777777778
<egg|tea|egg>
uh
<egg|tea|egg>
ah right
<egg|tea|egg>
;rpn 939 1054 * 3600 /
<kmath>
egg|tea|egg: 274.91833333333335
<egg|tea|egg>
bofh: so that's lower than your 330 W?
<egg|tea|egg>
though I guess it might have been doign something other than dissipating heat as well
<egg|tea|egg>
whitequark: was котя's blanket fort not effective?
<Iskierka>
was this at least a good chip to justify that heat output?
<bofh>
egg|tea|egg: so at leaat 20W are from HDD+cdrom,the rest of the difference I'll chalk up to manufacturing variability, error in my wattmeter and electrical energy both not converted to heat & heat absorbed by materials.
<egg|tea|egg>
blarg I should upgrade this machine to windows 10 someday, the lack of cuneiform fallback is annoying
<bofh>
egg|tea|egg: as for the context for that post? well I am merely thanking the local Atlantic sea/ocean deity for the relative calm weather currently :P
<egg|tea|egg>
bofh: I wasn't asking for context, I was merely providing some for my complaint about lack of cuneiform fallback on windows 7
<egg|tea|egg>
bofh: also I'm not entirely sure this is correct akkadian
<egg|tea|egg>
where's an akkadian linguist when you need one
<Fiora>
can't find it. maybe there wasn't one!
<Fiora>
and i just found a whole set of atelier cosplays I hadn't seen before.
<bofh>
egg|tea|egg: ahh. also sec, let me poke one.
<kmath>
https://twitter.com/bofh453/status/921870171844890624 <bofh453> "Ruthenium hexafluoride is made by a direct reaction of ruthenium metal in a gas stream of fluorine and argon at 400–450°C." AAAAAAAAAAAAAAA
<bofh>
I will say it's certainly a brisk synthesis
<egg|tea|egg>
bofh: you know a nice selection of people :-p
<kmath>
https://twitter.com/bofh453/status/921870489479467008 <bofh453> "Ruthenium hexafluoride is made by a direct reaction of ruthenium metal in a gas stream of fluorine and argon at 400-450°C." AAAAAAAAAAAAAAA
<Fiora>
egg|tea|egg: which i just RT'd
<egg|tea|egg>
nice
<egg|tea|egg>
Fiora: also I'm hopelessly confused by the english-french correspondences on the names of clothes, it's even worse than metallurgy
<kmath>
<eggleroy> (this joke was co-authored with @whitequark, both authors contributed equally to this work)
<bofh>
egg: that referenced tweet is still magnificent, btw.
<egg>
:D
<egg>
a bit abstruse (language barrier + set theory) sadly so few get it
<bofh>
hey, those are the best sorts of jokes :P
<egg>
Fiora: context for that french tweet co-authored with whitequark: in french ordered pair is "couple", which also means a couple in the "people in a relationship" sense; unordered pair is "paire". An ordered pair can be constructed from set theory as a singleton and a pair, so that (a,b) := {{a}, {a,b}} (Kuratowski's construction)