914900+ entries in 0.644s
mjr___: it depends on whether you
think we need constant improvement
KRS-1: Diablo: We are getting more clever at improving what we do, but we are not changing how we do it. Example: ever smaller
tolerances for CPU design. At some point moore's law will have
to break down and not apply. Do you
think we will enter a sort of "dark ages" where
there will be little, if any discovery for a very long
time? Or, do you
think
there will be great discovery on
the
Diablo-D3: they both licensed existing IP from
their fabs and spammed it on
their chips with minimal routing and
timing wiring
joecool: bfl is an avalon with
the name ground off
mjr___: by
the way, is asicminer an actual chip or a hardcoded fpga?
mjr___: yeah, i was
talking about
the one from
taiwan semi
Diablo-D3: mjr___:
the korean one Im
talking about was much smaller
mjr___: yeah
the finfet operated on .7 volts at 28nm
Diablo-D3: the reason you decrease feature size is
to decrease resistance
through less voltage
KRS-1: ok so bad use of
the
term but i
think you know what im referring
to?
Diablo-D3: what sort of friction are we
talking about here?
KRS-1: no friction at
that scale?
KRS-1: ah and
thus a friction reduction
joecool: just swapping material, can maintain
the design
KRS-1: but nanotube
transistors is a deviation from
traditional cpu design isnt it
Diablo-D3: and some research
team from Korea has developed a working FinFET
KRS-1: nano=extremely small + billions of
transistors = CPU no?
KRS-1: that could
take us off on a completely different
tangent
Diablo-D3: IBM has produced a cabon nanotube
transitor
Diablo-D3: mjr___: and
the CUs set up
the arguments as shared registers for
the entire work group
Diablo-D3: mjr___: and btw, I only upload
the kernel arguments once every 2**32 hashes
KRS-1: lol never even heard abou
tit
Diablo-D3: thats not an issue I have
to care about
mjr___: and with pipelining,
the best you can hope for is one double hash per clock cycle
mjr___: you
take 256 clock cycles just
to get
the hash in over a serial bus
mjr___: Diablo-D3: i
think eventually we will need
to use a larger bus, serial is a bottleneck
Diablo-D3: joecool: 8nm is
the half node size for 10 iirc, so its
the same shit anyhow
Diablo-D3: and rotate is more expensive
than add and xor on nvidia
joecool: Diablo-D3: actually intel claimed
they might skip 10nm and go
to 8nm
Diablo-D3: nvidia does not do ANY of
those in a single cycle
KRS-1: I dont know much about hynix and
toshiba but I bet samsung is doing cool stuff. Hehe
thinking back, I was a very big fan of Motorola.
mjr___: Diablo-D3: do you
think you could implement
the sha-256 algo in a way
that would be efficient for nvidia?
Diablo-D3: joecool: it becomes impossible
to design chips at 10nm
joecool: Diablo-D3: quantum
tunnelling will be more of an issue at
the 10nm process level
Diablo-D3: KRS-1:
they're already working on 14nm processes
Diablo-D3: KRS-1: now, samsung, hynix, and
toshiba fab chips
too
Diablo-D3: they refuse
to optimize for integer performance
Diablo-D3: mjr___: yes, but
they stick with proven
technologies
KRS-1: yea i heard
that, but
they once did i believe
mjr___: amd does design
though don't
they?
Diablo-D3: they split
their fab division off as GloFo
KRS-1: who is right behind
them or are
they head and shoulders above
the competition?
Diablo-D3: also, intel is finishing construction on
their 14nm facility in arizona
Diablo-D3: so if you're designing
these you end up with much larger gaps between
the features
KRS-1: not sure if
the future is indeed bright or if we will eventually lead
to a dark age where
there is minimal discovery..that kind of bothers me, because we're only improving but not changing
Diablo-D3: well, its where
they can jump right
through material
KRS-1: thats where
the electrons dont know which path
to
take or whatnot
Diablo-D3: below
that, quantum
tunneling becomes an issue
Diablo-D3: KRS-1: with current lithography
techniques, chemical polishing, and
the other processes used
to make
these chips
KRS-1: anyway i'll read up on it-
thanks!
KRS-1: thats where
the chip design is actually carved out? I
thought it was on
these big chalk board looking
things
that are shrunk
to
the wafers before
the lithography
Diablo-D3: KRS-1: its not really a board, its a very
thin wafer
KRS-1: they must be massive on
the actual design boards
mjr___: is
that
the entire chip is being used, and most chips rely on
the fact
that some parts can cool while others are hot
KRS-1: which is
the printing board for cpu's right?
mjr___: that is
the hardest
thing about bitcoin mining
KRS-1: and managing
the heat
Diablo-D3: we're reaching
the limits of ultraviolet lithography
techniques
KRS-1: but yea..thats
the hard part.
KRS-1: lol
thats what SHE said
Diablo-D3: KRS-1: now
the
thing is, its not
the size of
the particle but
the size of
the pipe
mjr___: but still
traditional chips
mjr___: they were
talking about smaller and smaller processes for chips
mjr___: and yes quantum computing does exist right now...
they've gotten 7 qubits
Diablo-D3: KRS-1: since quantum space is quantized, we know what
the grid size for photons are
mjr___: not quantum in
this convo
truff1es: back
to
this subject again,quantum computing is far off in
the future dont worry
KRS-1: that is where i
think we will be going when moores law doesnt apply anymore.
KRS-1: anyway
the future of computing must involve
these principles at some point unless we will enter dark ages..no significant discoveries..that would be sad.
Diablo-D3: it DOES however have a limit of how close you can get
to it
Diablo-D3: KRS-1: well, it has no mass because it moves at
the speed of light </high school physics knowledge rape>
KRS-1: physics does get very weird in
the quantums.
KRS-1: they're maybe "in between" mass and energy
then..since mass IS energy and vice versa
mircea_popescu: but yes,
they probably don't have a size in
the classical sense
Diablo-D3: mircea_popescu:
traditionally
this is considered
true
KRS-1: ah
there we go...so photons do have a size but no mass-
that makes much more sense.
tiberiusiv: why
the fuck arent people in here outraged at mirceas second massive monthly loss
KRS-1: entanglement showed us
that
the primary rules of physics are secondary
to entanglements..distance and
time don't matter
Diablo-D3: anyhow,
the size of an electron is about 3 fermi
KRS-1: some very exciting
things are happening with light and quantum mechanics
KRS-1: thats what i was
thinking
Diablo-D3: although you'd have
to figure out how
to measure
the rest mass of a photon
KRS-1: do you
think
the future of computing is light
KRS-1: but
they have mass no?
KRS-1: they say we'd eventually come down
to
tolerances close
to
the size of atoms, although not smaller
than an electron, we might have
to move
to photons or something, although i
think photons are bigger
than electrons
KRS-1: thats sick..never
thought i'd see
that day
mjr___: they can't
take
the
time
to plagiarize?
mjr___: srsly
though, if any of
these reporters were doing a research project for a freshman high school class,
they would fail
kakobrekla: but i
thought bitcoin is like hospital
mjr___: that is correct however, as of right now, i don't
think BFL consumes any power!
KRS-1: we all have our own interests, whatever
turns our crank, namean? I don't
think it necessarily makes you a loser if you are not part of
the elite or mainstream.
mjr___: Diablo-D3: i hope
to see a 28nm process chip soon