← | Results 501 ... 650 found in all logged channels for 'f:asc fpga'
(trilema) asciilifeform: you can pick up a textbook and write a dram controller for fpga from first principles - and it won't work. because, for starters, only a small number of output cells in the chip can function on both rising and falling edge of clock cycle (what 'ddr' means) and only xilinx's closed turd knows where they are in the routing fabric;
(trilema) asciilifeform: it is trickier than first appears, because 'i want the s00p3r-s33kr1t floor plan for $fpga' is turing-complete (that is, whether the trade was an honest one is not resolvable by machine)
(trilema) ascii_field: see also the fpga threads.
(trilema) asciilifeform: which orbits around deep-packet shenanigans and likes fpga
(trilema) asciilifeform: the not-so-secret here is that intel wants to bake fpga directly on x86 die
(trilema) asciilifeform: there was an outfit which sold something quite like this (fpga on sdram stick) and even one in the shape of a cpu, which could sit down in ordinary cpu socket on multisocket mb
(trilema) asciilifeform: (connection via ram slots is not, as it may appear, lunacy, but could be done with fpga, which would pretend to be a slice of sdram to two boxes at the same time, ignoring refresh cycles and managing locking somehow)
(trilema) asciilifeform: aha, as on early fpga.
(trilema) asciilifeform: (then you get: 'fpga' which doesn't give a damn about radiation, can be fabricated with 1940s equipment, etc)
(trilema) asciilifeform mined his first, and for long time, only bitcoin, on an fpga for which he wrote a very elementary miner, as an exercise
(trilema) ascii_field: from spec sheet: 800MHz sparc-like; sata-2; 12 ddr3 ecc dimm slots; various standard i/o (e.g., pci-e, GbE;) and, most interestingly, built-in fpga for 2d interconnect into clusters by directly linking bus (202 G/s claimed.)
(trilema) asciilifeform: for another, fpga and the genuine article can never be comparable on account of ruinous delays in the latter
(trilema) asciilifeform: mircea_popescu: possibly the fpga/dragonfly thread relates.
(trilema) asciilifeform: (afaik no one has even produced a true (non-metalloidfpga) asic yet)
(trilema) asciilifeform: mention of xilinx is especially on the point because a bit-flip in fpga routing fabric is catastrophic
(trilema) ascii_lander: ben_vulpes: http://www.rocketlabusa.com/about-us/electronic-systems/avionics/ << radhard fpga is available, for its weight in diamonds, but does anyone actually fly with d-sub connectors? they aren't rated for vibration...
(trilema) asciilifeform: it had a xilinx spartan 1st ed. fpga inside. and was built like a tank.
(trilema) ascii_field: gabriel_laddel: not exactly. even if you could secure an infinite supply of the particular fpga, it turns out that they are not actually very general. in that you cannot efficiently implement muller c-gates in them.
(trilema) ascii_field: at one time there was much crowing re: a 'xilinx fpga backdoor'
(trilema) ascii_field: (for pathologically degenerate case, see my fpga tale from conf-II)
(trilema) asciilifeform: or my mips-on-fpga that exists nowhere but my living room
(trilema) asciilifeform: and especially not if they had used actual ab initio fab instead of 'hard copy fpga'
(trilema) ascii_field: on any fpga whatsoever
(trilema) ascii_field: mats: the way the story normally ends is that the reversed fpga becomes quasi-usable nearly the same time it goes out of print and replaced with incompatible version...
(trilema) asciilifeform: kakobrekla: recall my fpga story at conf. II ? same idea
(trilema) asciilifeform: if i'm buying 'fpga', and he doesn't have the exact bitstream, he's more or less shit out of luck
(trilema) asciilifeform: sucks for exactly the same reasons as ordinary fpga
(trilema) asciilifeform: called antifuse fpga
(trilema) asciilifeform: but in practice, it has. i have a $$$$$ fpga board gathering dust because realized that it is impossible to implement ddr2 controller (and drive any available memory of reasonable density) without reversing the entire chip
(trilema) asciilifeform: mircea_popescu: an fpga that isn't a usg-proprietary piece of shit would go a long way
(trilema) asciilifeform: it's a little thing with a 'xilinx' fpga (nonvolatile) and 'ftdi' usb2 diddler
(trilema) asciilifeform: theer is an 'open bench logic sniffer' (more or less naked fpga, ~50 usd)
(trilema) asciilifeform: interestingly, fpga dev boards often have bayonet connectors for external clocking.
(trilema) asciilifeform: the mere emission of a 'muhahahaha' - or some more specific activity peculiar to reverse-engineering purloined fpga bitstreams ?
(trilema) asciilifeform: 'PARASTOO IS INFORMING THAT "CPLD AND FPGA BITSTREAMS" RUNNING THE SECURITY MODULES IN BOTH GROUND STATIONS AND MANY OF THE HOVERING ACTUAL SATELLITES THAT ARE LINKED TO SOME TARGETED NOCs ARE FULLY DECRYPTED ..AND WE FOUND -- WHILE DOING SATANIC LAUGHS -- SOME OF THESE UGLY CODES ARE NOT EVEN COMPATIBLE WITH BASIC NIST'S FIPS SECURITY GRADE GIVEN TO THESE PRODUCTS WITH $+APPLAUSE .'
(trilema) asciilifeform: i still want that fpga that evolves to live in one particular floor tile of your cellar.
(trilema) asciilifeform: folks doing actual work with fpga make very light use of the general-purpose routing matrix
(trilema) asciilifeform: this is because the entire fpga concept, as imagined by n00bs to the subject, does not actually work.
(trilema) asciilifeform: punkman: [fpga thread] would MIPS be easier than x86? or does the problem lie in connecting the thing to outside world (fast RAM, peripherals, etc) << problem is that you can't make so much as a barrel shifter on any extant fpga without using a closed turd from the vendor - actually, two turds: one in the logic library part itself, other - the synthesis toolchain
(trilema) ascii_modem: folks who bring up fpga as a solution - straight to ebay, pick up a board, see if you can bake so much as an i386 compatible, with at least the original's performance, -without- the fpga vendor's closed libraries
(trilema) asciilifeform: fpga is internally very much like the well-known 'eniac'
(trilema) asciilifeform: thing is, synthesis for proper asic is nothing like fpga synthesis
(trilema) asciilifeform: anyone who wants one - can go get (it builds for the larger among the common fpga)
(trilema) asciilifeform: the latter isn't simply a non-reprogrammable fpga; it has the LUTs of fpga, but with 'wiring' baked into the metallization layers, rather than fpga-style switching fabric
(trilema) asciilifeform: (an undergrad can throw a mips-compatible fpga core together in a day)
(trilema) asciilifeform: !s hard copy fpga
(trilema) asciilifeform: also to prevent taking the design to another fpga.
(trilema) asciilifeform: at the risk of repeating the last 100+ xilinx threads - the closed architecture of -all- fpga vendors is specifically to enable this 'business model'
(trilema) asciilifeform: incidentally, fpga work shits straight into the faces of folks who think that 'anything can be slow-prototyped'
(trilema) asciilifeform: decimation: overpowers << skeptical until demonstrated in field (i.e. x86 pc) rather than laboratory (fpga with custom dram controller)
(trilema) asciilifeform: the way i understand it, a 100% open fpga, produced in a thousand factories of mutually-antagonistic owners, is the one and only way to 'open computer.'
(trilema) asciilifeform: the closest thing anyone has conceived off to movable type re: chip design is the fpga concept.
(trilema) asciilifeform: wasn't speaking of fpga specifically
(trilema) asciilifeform: if you had cheap chip fab, there would not even be much reason to build fpga !
(trilema) asciilifeform: decimation: notice that all known fpga manufacturers (xilinx, altera, lattice, a few others) have the same business model
(trilema) asciilifeform: don't wait for the chinese to copy fpga. they have no reason whatsoever to.
(trilema) asciilifeform: as i may have once explained, the chicoms have not done this and should not be expected to ever do it, because fpga is profoundly useless in a mass consumer product.
(trilema) asciilifeform: pci is a paragon of transparency compared with any extant fpga
(trilema) asciilifeform: where is the open fpga?
(trilema) asciilifeform: (re: miner asics: anyone who gives a damn can find, in #b-a logs, my reasonably well-supported hypothesis that miner asic never actually -happened.- that is, there are devices, and they - approximately - work, but they are not 'asics' in the traditional sense. more 'hardcopy fpga.' - actual term of art)
(trilema) asciilifeform: mircea_popescu: not only dreamed, but simulated using bizarre terrorist contraption of fpga+dram+flash.
(trilema) asciilifeform: mircea_popescu: cannot, much as i'd love to, use 486, or bk-0011, or handmade pdp-11 in fpga.
(trilema) asciilifeform: on fpga emulating a 486?
(trilema) asciilifeform: e.g., xilinx wants you to prototype on their fpga, and then offers massively-discounted asic process which simply consists of their fpga die plus custom metallization layer.
(trilema) asciilifeform: decimation: asic << and here's where the turdmeisters win. many 'asic' products on the market presently are actually 'hardcopy fpga'
(trilema) asciilifeform: this - is fpga 'in a page.'
(trilema) asciilifeform: which makes the entire proposition of fpga a thinly-veiled scam
(trilema) asciilifeform: decimation: what joy. more proprietary crud that has to be on the die for fpga to talk to anything external.
(trilema) asciilifeform: mircea_popescu: remember the self-evolving fpga?
(trilema) asciilifeform: all i really need in terms of hardware - is an fpga which isn't a turd.
(trilema) asciilifeform refrains from launching into rant re: how you can actually fit 10x more gates in fpga if you start 'cheating' by using secret data re: the internal layout, and how any system which offers this capability is actually telling lies when advertised with its highest-level intended abstractions as the correct ones.
(trilema) asciilifeform: 'aha, block xxxxx, this is when i tried mining on fpga'
(trilema) asciilifeform: decimation: yes. incidentally this is why we can't really do asynchronous logic (e.g. muller's gate) on extant fpga.
(trilema) ascii_on_tour: !s fpga evolve
(trilema) asciilifeform: the kind of machine that absolutely requires an fpga - simply has not been designed yet.
(trilema) asciilifeform: decimation: have to understand what fpga is used for
(trilema) asciilifeform: shame fpga was invented too late for soviet version
(trilema) asciilifeform: decimation: problem is lack of fpga with documented layout, so you aren't chained to a vendor's turdware.
(trilema) asciilifeform: thing you must appreciate, if never having worked with fpga, is that they just barely work.
(trilema) asciilifeform: because fpga in the schoolbook sense is not actually available.
(trilema) asciilifeform: basically, fpga works, but all existing high-density examples are sold by scammers of a kind - who absolutely could not abide the buyer using these items as the schoolbook promised - 'reconfigurable computer'
(trilema) asciilifeform: the serious fpga market (vs. 'lattice' & friends, who largely get used for minor glue logic) is a duopoly.
(trilema) asciilifeform: (x86 is rather too complex to actually implement in any extant fpga and still leave room for something useful on top)
(trilema) asciilifeform: because 'xilinx' sells, by the tonne, a large fpga with built in 'ppc' cores.
(trilema) asciilifeform: not all that similar. cpu in fpga fabric. you can do this right now
(trilema) asciilifeform: no need to solder, surplus fpga demo board off 'ebay'.
(trilema) asciilifeform: bounce: if you want this (as i do) built it by hand, with fpga-to-pcie board.
(trilema) asciilifeform: to make long story short, there are exactly two vendors of 'high-density' fpga, and their products are almost exactly alike - more so even than 'coca cola' and 'pepsi.'
(trilema) asciilifeform: it is virtually impossible now to buy a pure fpga.
(trilema) asciilifeform: what can one accomplish on fpga, knowing nothing about your opponent's future purpose for it, with a small turd ?
(trilema) asciilifeform: an unexpected cpu in the middle of fpga LUT fabric would stand out even under optical microscope.
(trilema) asciilifeform: if satan is merely listening to your i/o, this is a listening device in the ordinary sense, and not really fpga-specific diddling
(trilema) asciilifeform: let's say you purchase an fpga from satan.
(trilema) asciilifeform: fpga is still interesting
(trilema) asciilifeform: bounce: nope. most fpga work is carried out in units considerably larger than 'gate', from considerations of efficiency.
(trilema) asciilifeform: bounce: all the extant fpga have tightly closed architecture
(trilema) asciilifeform: fpga mining board (as late as 2013) advertised by 'cardreaderfactory', a firm which appears to sell 'card skimmers' (a type of cheating device, popular in usa, where credit card magstripes are secretly read from unsuspecting victim)
(trilema) asciilifeform: "brainfuck" machine on fpga >> nope. (why not? >> http://www.loper-os.org/?p=55)
(trilema) asciilifeform: likely contents of sd: fpga bitstreams
(trilema) asciilifeform: no one seriously cares what was in the fpga miners from 2 yrs. ago
(trilema) asciilifeform: both of the major (well, only ones worth mentioning) fpga vendors make most of their profit from renting out license to use their libraries (ethernet controller, soft cpu cores, etc.) and not from hardware. hence the tightly closed architectures.
(trilema) asciilifeform: (actually there may be. fpga/cpu combo is a dime a dozen - xilinx made ppc/fpga cores for ages. but a modern pc arch cpu is something rather else)
(trilema) asciilifeform: afaik there is no mass-market arm/fpga combo ic
(trilema) asciilifeform: but at least these were traditional fpga and one could (at least in the case of xilinx) use linux toolchain, traditional vhdl/verilog, etc
(trilema) asciilifeform: decimation: for some years, there were fpga on the market that fit into an intel or amd cpu socket
(trilema) asciilifeform: decimation: these are a dime a dozen. not very useful for fpga work, as the latter have proprietary pieces that, if you do not use, leave you with a monstrously bloated design that fits nowhere.
(trilema) asciilifeform: and build the actual state in fpga out of same.
(trilema) asciilifeform: decimation: think 'emulation of room-sized fpga' and you're 90 percent there.
(trilema) asciilifeform: if it was one of the (rare) fpga clusters, it might've been interesting.
(trilema) asciilifeform: i did get a number of people playing with fpga (the closest thing to 'freedom' in this context, that i know of) who perhaps wouldn't have otherwise.
(trilema) asciilifeform: jurov: my half-baked fpga prototype cost me about 500 usd. but picture, in my shoes, proposing that it be manufactured en masse, to an industrial magnate.
(trilema) asciilifeform: decimation: the most basic 'jungle fpga' is just your ordinary ROM.
(trilema) asciilifeform: (XC6216 was the one and only known fpga where random wiring was guaranteed -never- to release the magic smoke.)
(trilema) asciilifeform: BingoBoingo: vaxberry pie << pick up fpga dev board on ebay.
(trilema) asciilifeform: perhaps if cray were alive, you could buy 'fpga' the size of a billboard...
(trilema) asciilifeform: and that, while certain interesting classes of attack are impossible on an fpga,
(trilema) asciilifeform: my point was that you end up trusting a big fat american duopoly when using modern fpga.
(trilema) asciilifeform: (fpga has no idea where the 'cache' will be on the fabric, or even that it will be configured as cpu rather than, say, decompressor)
(trilema) asciilifeform: it is quite true that you cannot put this kind of trap in an fpga without making the whole thing look quite ridiculous even under an optical microscope.
(trilema) asciilifeform: several places have massive fpga clusters, hijack those.
(trilema) asciilifeform: if fpga with published internal specs existed (as it did in 1994, say) but of modern proportions, this could change.
(trilema) asciilifeform: sort of how i once pontificated re: fpga implementation of bitcoind (e.g. verilog) would be a good 'cruft shedder' exercise.
(trilema) asciilifeform: in the early days of fpga mining
(trilema) asciilifeform: BingoBoingo: plenty of people own fpgas. me, for instance.
(trilema) asciilifeform: today there are fpgas with several mil. of gates, costing in the high five figures (u.s. dollar)
(trilema) asciilifeform: interestingly, the earliest (late '80s) fpgas did not have this problem
(trilema) asciilifeform: likewise, each vendor (and even particular flavours of fpga from same vendor) provides slightly different building blocks on top of the usual LUTs
(trilema) asciilifeform: if you don't optimize for a particular fpga, you can't get anything close to the maximum rated clock of the sdram in question
(trilema) asciilifeform: because the opencores libs aren't optimized for a particular fpga
(trilema) asciilifeform: the correspondence between each bit of the bitstream and the internal logic of the fpga is a $billion secret
(trilema) asciilifeform: or hell, even fpga with sdram controller
(trilema) asciilifeform: likewise, and even more elementary: can't make a self-reconfiguring xilinx (or other fpga based) system without x86/unix in the loop unless you can ditch the vendor's toolchain
(trilema) asciilifeform: mircea_popescu: if you're sitting on a warehouse full of fpga.
(trilema) asciilifeform: mircea_popescu: i've no incentive to do this, but the folks presently stoking their furnaces with old fpgas - do.
(trilema) asciilifeform: trick is to devise an alt where the ideal miner is actually a full fpga.
(trilema) asciilifeform: re: the folks sitting on heaps of obsolete fpga: why hasn't even one started an altcoin with a diddled workfunction
(trilema) asciilifeform: yeah actually even my own toy fpga miner had this one.
(trilema) asciilifeform: Luke-Jr: sdram plus fpga which does nothing but sCrypt, vs. a general-purpose pc?
(trilema) asciilifeform: everybody knows that you can get FPGA+sdram socket boards for pocket change, right?
(trilema) asciilifeform: question then is: ebay is chock-full of FPGA miner rigs being sold off for pennies on the dollar. why isn't anyone buying them up for use as space heaters?
(trilema) asciilifeform: I have an FPGA board right here on my desk, with a completely verifiable TCP/IP stack
(trilema) asciilifeform: the client would choose the fpga and the board, and could even physically supply it
(trilema) asciilifeform: the product I envision would be a single-purpose device which runs the block chain and only the block chain. the FPGA code implementing the instruction set, firmware, etc. would form a complete description of the system.
(trilema) asciilifeform: there are numerous available FPGAs which have a programming interface which is not accessible from the routing fabric (i.e. separate pins.)
(trilema) asciilifeform: and so, one can use any of a thousand commercial fpga dev boards
(trilema) asciilifeform: I'm thinking of a node (not a miner, note) written for a custom architecture, running on an fpga (of the client's choice)
(trilema) asciilifeform: FPGA is cheap you know
(trilema) asciilifeform: I am personally working on an FPGA setup for LTC
← |
Download hourly DB snapshot | Get Source Code