berndj | bart416, i thought czochralski rigs used silica crucibles (with the side benefit of introducing oxygen impurities). or what all this "you need high carbon steel" for? | 01:14 |
---|---|---|
berndj | lol @ refractory materials above 10kK | 01:15 |
bart416_ | <berndj> bart416, i thought czochralski rigs used silica crucibles (with the side benefit of introducing oxygen impurities). or what all this "you need high carbon steel" for? <-- was it a czochralski crucible? | 07:03 |
berndj | actually i'm not sure anymore what you guys were discussing; i thought it was CZ but when i looked again you might've been talking about making copper wire | 07:04 |
bart416 | Yeah it was confusing what sort of furnace he was talking about | 07:09 |
bart416 | What I do know is that CaO is the worst material ever to make a furnace | 07:09 |
bart416 | And it's the last you'd want to use to grow crystals | 07:10 |
B0101 | Hi azonenberg | 13:15 |
azonenberg | hi | 13:15 |
azonenberg | How goes the lab setup? | 13:15 |
B0101 | going well... though my neighbors are a little worried about me doing experiments in my own home | 13:16 |
azonenberg | Not surprising | 13:17 |
azonenberg | Just be up front about it and dont try to hide | 13:17 |
azonenberg | it looks better that way if anybody comes asking questions | 13:17 |
bart416_ | Are your neighbours constantly looking through your windows or something? | 13:20 |
B0101 | nope | 13:20 |
berndj | worried like grandmas worry about their grandkids or worried like they might call 911 on you | 13:20 |
B0101 | the former | 13:21 |
azonenberg | oh, then thats fine | 13:22 |
azonenberg | the second is what you want to be cautious about lol | 13:22 |
B0101 | well once, my friend and her child came over, but somehow, the child got into my room and was playing with my 50KV power supply | 13:25 |
bart416_ | Always lock away your lab equipment | 13:26 |
bart416_ | That's a basic rule | 13:26 |
azonenberg | bart416_: We just dont allow kids in the place, it'd be too hard to childproof given the random stuff we have lying around | 13:27 |
azonenberg | when i get a house of my own the lab will all be in dedicated rooms with auto-closing doors and card readers | 13:27 |
bart416_ | That's just paranoid :| | 13:27 |
azonenberg | bart416_: no, its good policy :p | 13:28 |
bart416_ | auto closing doors I'll agree on | 13:28 |
bart416_ | But card readers... | 13:28 |
azonenberg | all the labs at school do that | 13:28 |
azonenberg | card readers mean i can give friends cards rather than keys | 13:28 |
azonenberg | keep audit logs if something breaks | 13:28 |
bart416_ | you let other people use your equipment? | 13:28 |
azonenberg | i plan to construct a building with a living area and a lab space | 13:28 |
azonenberg | cleanly separated | 13:29 |
azonenberg | so for example friends staying over wont be allowed into the lab without someone present | 13:29 |
azonenberg | and if i have someone working on a project with me and i trust them enough they'd have 24/7 lab access but not be allowed into the living spaces under normal circumstances | 13:29 |
bart416_ | Heh, you don't have a plan for the house you want yet? | 13:29 |
azonenberg | bart416_: not floor plan level, that will be done based on the plot of land i pick | 13:30 |
azonenberg | but at list-of-requirements level, yes | 13:30 |
bart416_ | No ideas yet of what you want? | 13:30 |
azonenberg | You mean, at the floor plan? No | 13:30 |
azonenberg | I know i want a class 1000 cleanroom, a machine ship, an electronics lab, a dedicated server room | 13:31 |
azonenberg | conference room/media center on the border between the lab and the living area | 13:31 |
azonenberg | some sort of armory in the basement where i keep all of the stuff i go target shooting with | 13:32 |
azonenberg | That'd be a separate card list, just me | 13:32 |
B0101 | azonenberg: wow, you plan to build your own lab and house | 13:33 |
azonenberg | B0101: Its at least ten years out | 13:33 |
azonenberg | ~5 for the phd and that much again of workign full time to save up enough to build it | 13:33 |
azonenberg | But yes | 13:33 |
azonenberg | Then a SEM somewhere, probably in the cleanroom area | 13:34 |
B0101 | wow, here, u got to ask for permission even if you wanted to change the window | 13:34 |
azonenberg | B0101: I'd buy an empty plot of land (or one with an old building and tear it down) | 13:34 |
azonenberg | Get all of the necessary building permits | 13:34 |
cheater | azonenberg: you could also have guest areas with direct access to the work area, for people who want to be able ti live-in and work but don't want to bother with the social stuff. | 13:34 |
azonenberg | And design/build the whole place from scratch | 13:34 |
azonenberg | cheater: I might | 13:34 |
azonenberg | Like the media center for example | 13:35 |
cheater | eg being seen by your wife and kids with only their boxers on. | 13:35 |
azonenberg | which would double as a conference room for the lab | 13:35 |
azonenberg | And that would not be allowed | 13:35 |
azonenberg | dress code for the lab area is fully clothed plus lab coat if necessary | 13:35 |
azonenberg | or bunny suit, in the cleanroom | 13:35 |
B0101 | Azonenberg: in here... almost EVERY building is controlled by government | 13:35 |
azonenberg | if you arent wearing anything under the bunny suit, who cares | 13:35 |
cheater | it was just an example, but being unshaved or uncombed are other options | 13:35 |
azonenberg | oh, yeah | 13:35 |
azonenberg | B0101: Not in the good old usa, here you can own the building | 13:36 |
cheater | B0101: where is that? sounds like russia | 13:36 |
B0101 | Singapore | 13:36 |
cheater | actually no, russia has got those freedoms | 13:36 |
cheater | communist russia didn't | 13:36 |
cheater | singapore.. i feel with yo | 13:36 |
cheater | u | 13:36 |
azonenberg | cheater: If i build a guest-bedroom type facility i might put it on the border as well | 13:37 |
azonenberg | have two doors | 13:37 |
B0101 | cheater: the democratic communist state of Singapore (as some people call it) | 13:37 |
cheater | yes | 13:37 |
cheater | B0101: yes | 13:37 |
azonenberg | cheater: i'd probably have the shared area in the middle and put the front door there | 13:38 |
azonenberg | Both sides can get into it | 13:38 |
azonenberg | and enter/leave the building through it | 13:38 |
azonenberg | then you need a card to leave in either direction | 13:38 |
azonenberg | leave the common area* | 13:39 |
azonenberg | i.e. into the lab or the living spaces | 13:39 |
cheater | how about a chinese wire puzzle instead | 13:39 |
B0101 | hmm, biometrics or card system? | 13:39 |
azonenberg | lol | 13:39 |
azonenberg | B0101: Not fingerprints | 13:39 |
azonenberg | a) doesnt work with gloves on | 13:39 |
azonenberg | b) you leave them everywhere, easy to clone | 13:40 |
azonenberg | Retinal / iris scans are expensive | 13:40 |
cheater | you can clone them off the reader | 13:40 |
cheater | with gelatin | 13:40 |
azonenberg | if i can get them, i'll use them | 13:40 |
azonenberg | cheater: there are a lot of ways | 13:40 |
cheater | why | 13:40 |
azonenberg | point is, they arent that strong | 13:40 |
cheater | why do you need so much security :) | 13:40 |
azonenberg | cheater: its in part the convenience factor | 13:40 |
azonenberg | not having to carry around keys | 13:40 |
azonenberg | also, due diligence and liability | 13:40 |
azonenberg | significantly reduces the risk of a visitor getting hurt by stupidity | 13:41 |
cheater | what when the power is off | 13:41 |
azonenberg | since they cant get into anything dangerous unless i trust them enough to give them access | 13:41 |
azonenberg | cheater: then the doors revert to being key or combination locks | 13:41 |
azonenberg | i'd have a key in a safe in the common area | 13:41 |
azonenberg | that i could use in emergencies like that | 13:41 |
azonenberg | the point is, i dont have to give out keys | 13:41 |
B0101 | hmm, password protected lock... | 13:41 |
azonenberg | to let someone in | 13:41 |
azonenberg | Same reason hotels use cards | 13:42 |
azonenberg | easier to add/remove a serial number than to rekey a door | 13:42 |
azonenberg | Also, audit trail | 13:42 |
azonenberg | if someone goes into a lab and breaks something i know who was in there recently | 13:42 |
azonenberg | I just see it as being better for all involved | 13:43 |
B0101 | azonenberg: I have seen keypad locks with card readers | 13:43 |
azonenberg | B0101: I might do something like that too | 13:43 |
azonenberg | battery backup? | 13:44 |
azonenberg | so you can combo in if the card system is down? | 13:44 |
azonenberg | The only potential downside would be privacy but "john doe was here at 2 AM and used the electron microscope for 45 minutes, then left" is not something most people would care if the world knew | 13:44 |
B0101 | Azonenberg: the keypad and card lock I'm talking about requires you to scan your card then enter a password | 13:45 |
azonenberg | B0101: Combination is probably overkill | 13:45 |
azonenberg | if i did two factor it'd be card + biometric or combo + biometric | 13:45 |
azonenberg | cheater: also, i could integrate it with a burglar alarm (since the equipment would obviously be expensive) | 13:46 |
azonenberg | if you key in you need to disarm the alarm at the keypad | 13:46 |
azonenberg | but if you card in, it does both in one step | 13:46 |
B0101 | O.o | 13:47 |
B0101 | now people want to see my lab??? | 13:47 |
azonenberg | "people"? | 13:47 |
azonenberg | neighbors or police? lol | 13:48 |
B0101 | the people i'm talking about are well: my friends | 13:48 |
azonenberg | oh, lol | 13:48 |
azonenberg | Once people find out i have a setup this nice they usually want to see it | 13:48 |
azonenberg | Get used to it :p | 13:48 |
azonenberg | idk if you've seen my lab lately but http://imgur.com/a/Bnwst is the current state (last night) | 13:48 |
azonenberg | a couple of experiments in progress so not the most tidy on the workbenches | 13:49 |
B0101 | looks like my current eletronics work area, without the chemicals | 13:50 |
azonenberg | Three sit-down workbenches (for soldering/wet work, electronics testing/assembly, and microscopy) | 13:50 |
azonenberg | and two standing (for spin coating, baking, and contact lithography) | 13:51 |
azonenberg | then the server rack | 13:51 |
B0101 | azonenberg: well you have a great lab | 13:51 |
azonenberg | thx | 13:51 |
azonenberg | Not sure if you can see in the pics but we have full fire sprinkler coverage | 13:51 |
azonenberg | the entire building has a dry-pipe system | 13:51 |
azonenberg | The binder on the vertical shelf next to the safety glasses is a MSDS book | 13:52 |
B0101 | oh | 13:52 |
azonenberg | *prints out msds for diamond polishing compound* cant forget that | 13:53 |
B0101 | anyone fancies a "warning: trespassers with me used for experiments" sign? | 13:53 |
azonenberg | lol | 13:53 |
azonenberg | i've seen that | 13:54 |
azonenberg | Might not go over too well with the neighbors :p | 13:54 |
B0101 | If i told them my neighbors that I bought uranium, i'm sure that they will kill me | 13:55 |
azonenberg | lol | 13:55 |
azonenberg | http://unitednuclear.com/images/lasersign.jpg | 13:56 |
azonenberg | http://unitednuclear.com/images/sign3.jpg | 13:57 |
B0101 | In this country, a little experiment of everyone near you, and i mean EVERYONE near you | 13:58 |
B0101 | *will alarm | 13:58 |
B0101 | "danger: big scary laser..." | 14:00 |
azonenberg | Do not look into beam *with remaining eye* | 14:01 |
azonenberg | lol | 14:01 |
B0101 | lol | 14:01 |
Action: B0101 pictures a scene of doing an experiment and horror music playing in the background | 14:02 | |
azonenberg | Gotta be a proper mad scientist | 14:03 |
azonenberg | I havent seen villagers with pitchforks outside though - must not be trying hard enough | 14:03 |
Action: azonenberg needs more evil minions :P | 14:03 | |
B0101 | hehehe... | 14:03 |
azonenberg | And an army of killer robots :p | 14:06 |
B0101 | muahahahaha... | 14:06 |
B0101 | ok, this is going to scare children | 14:06 |
azonenberg | actually, in all seriousness i am working on an autonomous robot rubber-band-gun turret with a friend | 14:07 |
azonenberg | she claims to be one of my evil minions lol | 14:07 |
azonenberg | But i need more of them | 14:07 |
azonenberg | i dont have anybody who knows fiberglass or composites yet | 14:08 |
Action: B0101 wonder what can be with done some carbon | 14:10 | |
B0101 | azonenberg: heard of carbon nanotube transistors? | 14:12 |
azonenberg | Yes | 14:12 |
azonenberg | A little complex for amateur work though | 14:12 |
azonenberg | afaik you basically need a SEM to align everything | 14:12 |
B0101 | well, i'm going to bed, ttyl | 14:13 |
azonenberg | k | 14:13 |
bart416_ | Why not make those single atom transistors if you're going to switch to that sort of thing? | 14:16 |
azonenberg | lol | 14:30 |
azonenberg | Later this afternoon i am going to try and figure out what is up with my metal hardmasks not working | 14:30 |
bart416_ | btw, I have a nice toy arriving next week (though it's not for me) | 14:30 |
bart416_ | An industrial CO2 laser xD | 14:30 |
azonenberg | etch off all of the metal and see if the KOH still doesnt etch the silicon | 14:30 |
azonenberg | and lol, for what? | 14:30 |
bart416_ | Dunno, a friend asked me to buy one from amazon :S | 14:31 |
azonenberg | lol | 14:31 |
bart416_ | He's giving me 5% ontop for some reason :S | 14:31 |
Action: azonenberg makes mental note to wear long sleeves next time he welds | 14:31 | |
azonenberg | i still have a bunch of spatter marks on my arm from thursday's practice session lol | 14:32 |
bart416_ | (5% on 2000 euro is nice) | 14:32 |
bart416_ | When working with warm metal, wear thick fireproof clothing or no clothing at all... | 14:33 |
azonenberg | bart416_: i was wearing heavy jeans on my legs | 14:33 |
azonenberg | and i think a leather apron | 14:33 |
azonenberg | just forgot the lower arm | 14:33 |
azonenberg | everything from glove to shoulder got some spatter lol | 14:33 |
azonenberg | I'll remember next time lol | 14:35 |
azonenberg | gonna be down there a decent amoutn this semester because i want to get good at MiG and then learn TiG | 14:35 |
CIA-67 | homecmos r122 | trunk/lithography-tests/labnotes/azonenberg_labnotes.txt | Yesterday's lab notes | 14:40 |
bart416_ | Jeans is actually fairly good | 14:42 |
bart416_ | As long as it are only small spats that is | 14:42 |
azonenberg | Yeah, and leather boots | 14:43 |
azonenberg | like i said just a stupid oversight on my part | 14:43 |
azonenberg | i've welded before, just was never very good at it | 14:43 |
azonenberg | this was my first time on an auto-darking helmet | 14:43 |
azonenberg | its like night and day vs the always-shade-12 ones :p | 14:43 |
azonenberg | i could actually see where the torch was pointing before striking the arc | 14:44 |
bart416_ | Maybe you should buy one of these: http://www.modernapparels.com/fire_proximity_suit.htm | 14:45 |
bart416_ | They're annoying as hell to wear | 14:45 |
azonenberg | Is this from experience? | 14:45 |
bart416_ | Yes | 14:45 |
azonenberg | Lol, hwat occasion did you have to wear one | 14:45 |
bart416_ | NDA :( | 14:45 |
azonenberg | lol | 14:45 |
azonenberg | How about http://www.modernapparels.com/fire-entry-suit.html | 14:45 |
bart416_ | I suppose it's clear we were melting small batches of metal | 14:46 |
bart416_ | lol | 14:46 |
azonenberg | Not really clear, the same suit can be used for close range firefighting | 14:46 |
bart416_ | If you ever have to wear one, be sure to have something to pad your shoulders ;) | 14:47 |
azonenberg | lol | 14:47 |
azonenberg | Hopefully i never do | 14:47 |
bart416_ | That's my only suggestion | 14:47 |
bart416_ | The pants are weirdly enough comfortable | 14:48 |
bart416_ | It's the vest that's the problem | 14:48 |
bart416_ | Well, I suppose it's still better than those throw away overalls | 14:50 |
azonenberg | lol | 14:50 |
azonenberg | i wear a disposable tyvek lab coat for splash protection when doing etching | 14:50 |
azonenberg | gets annoying | 14:50 |
azonenberg | But weighs practically nothing | 14:50 |
azonenberg | i need to get myself a real cotton lab coat at some point | 14:50 |
bart416_ | We were fairly lucky in that aspect, we got to wear scrubs under our throw away overalls | 14:51 |
azonenberg | what was that for? Or can you not say | 14:51 |
bart416_ | To avoid contaminating our own clothes | 14:52 |
bart416_ | We were working with dangerous chemicals in that lab where I did my internship | 14:53 |
azonenberg | Makes sense | 14:53 |
bart416_ | Not to mention it's easier to remove scrubs instead of regular clothes | 14:53 |
azonenberg | Yeah | 14:53 |
bart416_ | Incase of accidents | 14:53 |
azonenberg | When you say "dangerous" do you mean corrosive? toxic? both? | 14:54 |
azonenberg | biohazard? | 14:54 |
bart416_ | Among those yes | 14:54 |
berndj | re retinal / iris scans, am i the only person in the world who remembers seeing _Demolition Man_ ? | 15:52 |
bart416_ | No | 15:52 |
bart416_ | Demolition Man was an awful movie | 15:53 |
bart416_ | It praises uncivilised behaviour | 15:53 |
berndj | i was planning to test out my auto-darkening helmet a bit later azonenberg; glad to hear you foind it a night and day difference! | 15:55 |
berndj | of course, but it's only a matter of time before crooks start routinely poking eyes out and chopping fingers off | 15:55 |
berndj | the fact that it won't work doesn't console me! | 15:56 |
bart416_ | For a fingerprint scanner it'd sort of work... | 16:23 |
mrdata | so, would 99.999% pure silicon be good enough for homecmos? | 19:45 |
bart416 | What sort of impurities are we talking about? | 19:46 |
bart416 | If it's not a dopant it'll fall under crystal defects | 19:46 |
mrdata | oh, usually aluminum is the worst to remove | 19:47 |
bart416 | The latter will decrease the yield but shouldn't prevent succesful junctions | 19:47 |
bart416 | Al might be problematic | 19:47 |
azonenberg | bart416: For my current MEMS, its probably good enough | 19:48 |
azonenberg | i am not using it as a semiconductor, only a structural material | 19:48 |
azonenberg | Though i am relying on it being monocrystalline | 19:48 |
bart416 | yeah for MEMS and simple conduction it should do fine | 19:49 |
bart416 | If you're forming junctions aluminium might be problematic | 19:49 |
mrdata | how does it affect the result? | 19:49 |
bart416 | It shifts the levels of the energy bands in certain locations | 19:50 |
bart416 | Resulting in improper junction behaviour or schottky diode effect | 19:50 |
mrdata | oh, and that will screw with the device properties.. | 19:50 |
mrdata | i see | 19:50 |
bart416 | the latter might be more problematic actually | 19:51 |
mrdata | what sorts of devices would be less sensitive to that? | 19:52 |
mrdata | i know solar cell applications are okay | 19:53 |
mrdata | but, what else would be relatively unaffected? | 19:53 |
Action: mrdata imagines making an array of devices, that differ slightly in their properties due to impurities; and recording test results for later use | 19:58 | |
mrdata | say, if each one is an artificial neuron | 19:59 |
bart416 | MEMS | 20:00 |
bart416 | well, some of them | 20:00 |
mrdata | ok | 20:00 |
bart416 | as long as you don't need junctions MEMS should do | 20:00 |
bart416 | Might lead to some capacitive defects but nothing you can't compensate for | 20:00 |
mrdata | and junctions would be the most problematic? | 20:01 |
bart416 | At a junction you want perfection | 20:01 |
mrdata | can i take a probablistic approach to making junctions? eg: mark as bad, the ones that dont work, in an array? | 20:03 |
bart416 | yeah, what do you think semiconductor manufacturers do? | 20:04 |
bart416 | Making computer processors is a game of statistics | 20:04 |
bart416 | But the lower your purity, the lower the yield | 20:04 |
mrdata | so, they make redundancy and test and re-route? | 20:04 |
bart416 | they don't reroute | 20:05 |
bart416 | They just make large enough batches and refine the process to maximise yield | 20:06 |
mrdata | so do they typically throw out a lot of stuff? | 20:07 |
mrdata | eg: half? | 20:07 |
mrdata | or more | 20:07 |
bart416 | Depends on the complexity of the production, the process, purity of materials | 20:09 |
bart416 | 50% won't be enough to run a profit at consumer prices I guess | 20:09 |
bart416 | maybe good for an asic run of a low quantity IC | 20:11 |
mrdata | i hear that for small parts, like NPN transistors, they are made in batches, and each one tested, and then categorized as 2n2222 vs 2n3904 for example | 20:11 |
mrdata | so, everything in the same bin has a specified range of properties | 20:13 |
bart416 | Yeah, but some are also defective | 20:14 |
bart416 | + NPN are two junctions | 20:14 |
bart416 | fairly simple design | 20:15 |
mrdata | yeah | 20:15 |
berndj | or sometimes they sell the ones with defects in the cache as "celeron" | 23:17 |
berndj | and 2n3055 is what it gets called if it doesn't make spec for anything else | 23:17 |
berndj | dammit, why do i answer first then notice they've quite :( | 23:18 |
azonenberg | Or if the error is in logic circuitry | 23:18 |
azonenberg | they sell a quad core as a triple :p | 23:18 |
berndj | oh yes, i should turn the time machine dial back to 2010s | 23:19 |
berndj | but, i thought logic is just a miniscule part of CPUs these days? that almost all of the die area is memory of one sort or another? | 23:19 |
XgF | L2/L3 Cache is about half of the die these days | 23:20 |
azonenberg | berndj: Cache is a big chunk | 23:20 |
azonenberg | But depends on the part | 23:20 |
berndj | register file is surprisingly big, due to the number of ports it has | 23:20 |
azonenberg | for example, gtx 480 has 15 of 16 shaders active | 23:20 |
azonenberg | and i'm too familiar with that lol | 23:21 |
azonenberg | mips register file in a simple 5-stage pipelined implementation is triple ported, two read and one write | 23:21 |
azonenberg | i have five ports (one write, two execute-stage read, two decode-stage read) in my 8-bit arch | 23:21 |
azonenberg | trying to build a 5-port 16-element 8-bit register file | 23:22 |
azonenberg | out of 16x1 bit dual port ram | 23:22 |
azonenberg | now that was fun :p | 23:22 |
berndj | isn't there an O(2^n) [n = # of ports] size factor? | 23:22 |
azonenberg | You can do linear in number of ports | 23:22 |
berndj | or is it not quite as bad as that | 23:22 |
azonenberg | if you only have one write port | 23:22 |
azonenberg | Make N copies of the array, where N is the number of independent read ports | 23:23 |
azonenberg | writes go to everything | 23:23 |
azonenberg | then each read port addresses its own independent image of the memory | 23:23 |
azonenberg | thats what i did in both my triple and 5-ported reg files | 23:23 |
berndj | yeah, i guess i was thinking of n-write-ports files | 23:23 |
azonenberg | oh, n-writes? | 23:23 |
Action: azonenberg shudders | 23:23 | |
azonenberg | no thanks | 23:23 |
berndj | indeed | 23:23 |
azonenberg | But in a classic pipelined cpu architecture, you are reading all over the place | 23:23 |
azonenberg | but only writing during the writeback stage | 23:23 |
berndj | well, pentiums have for years been retiring at least two writes per clock | 23:24 |
azonenberg | x86 is a horse of a different color | 23:24 |
azonenberg | i said classic | 23:24 |
berndj | afaik & iirc & ianal etc | 23:24 |
azonenberg | as in simple risc | 23:24 |
berndj | true enough | 23:24 |
azonenberg | not multiplie-issue superscalar | 23:24 |
azonenberg | my goal in these designs is minimal gate count and relatively high clock frequencies | 23:25 |
berndj | x86 is why there are now more transistors than ants | 23:25 |
berndj | last i looked there was a ia64 with > 1 gigatransistor | 23:25 |
berndj | mostly cache though | 23:26 |
azonenberg | Woudlnt surprise me, gpus are up there already | 23:26 |
azonenberg | Let me put it this way, my 8-bit arch uses 122 flipflops, 545 LUTs, and 312 spartan-3a slices | 23:26 |
berndj | (probably rather ARM's fault rather than intel's - just due to number of cores out there) | 23:26 |
azonenberg | which is under 20% of a 200k gate spartan-3a (this is equivalent to about 35k gates) | 23:27 |
azonenberg | Clocks at upwards of 100 mhz in a spartan-3a -5 speed fpga | 23:27 |
berndj | wow, you could have built that in 1979 then! | 23:27 |
azonenberg | 16x 8-bit registers | 23:27 |
berndj | (but not at 100MHz) | 23:27 |
azonenberg | Lol | 23:27 |
azonenberg | 8-bit memory address space (i.e. 256 bytes) | 23:27 |
azonenberg | and 16-bit IO address space using pairs of registers | 23:28 |
berndj | do you think we'll ever run out of 64-bit address space? | 23:28 |
azonenberg | The idea was a design that could fit *anywhere* | 23:28 |
azonenberg | even in the smallest fpga xilinx still makes | 23:28 |
azonenberg | and no, i do not | 23:28 |
azonenberg | We may eventually have >2^64 bytes of memory | 23:28 |
azonenberg | per system | 23:28 |
azonenberg | but we will be using distributed memory message passing architectures | 23:28 |
azonenberg | 2^32 per core is a little cramped but 2^48 is probably enouhg | 23:29 |
berndj | i wouldn't either, but # bits is roughly linear with time, according to moore's law | 23:29 |
azonenberg | by the time you need 256GB of memory you should not be using an SMP anymore | 23:29 |
azonenberg | it doesnt scale | 23:29 |
berndj | so 64 bits should last (at least) twice as long as 32 bits did | 23:29 |
azonenberg | look at blue gene for an extreme example | 23:29 |
azonenberg | my school's BG/L has 32768 compute nodes each with 512MB of 1GB of memory | 23:29 |
azonenberg | or 1GB* | 23:30 |
azonenberg | these are 32 bit procs | 23:30 |
azonenberg | addressing i think 12T of total ram | 23:30 |
berndj | nice toy! | 23:30 |
azonenberg | Because its not all on one compute node | 23:30 |
berndj | can it play tic tac toe? | 23:30 |
azonenberg | So thats where i see things heading | 23:30 |
azonenberg | lol, no idea - all i've used it for is matrix multiplication during the class :p | 23:30 |
berndj | yeah, NUMA ftw | 23:30 |
azonenberg | all of my HPC since then has been on my cuda cluster | 23:30 |
azonenberg | My current preferred arch for HPC is a NUMA system consisting of a bunch of SMP systems with a relatively fast multicore CPU (i7 or similar) and one or more GPUs each | 23:31 |
azonenberg | connected by sockets | 23:31 |
azonenberg | this is a deeper hierarchy than needed for some project | 23:32 |
azonenberg | sometimes a bunch of powerpc over MPI is all you need | 23:32 |
azonenberg | iow, bg/l or successors | 23:32 |
azonenberg | In any case, if you are designing a reasonable architecture | 23:33 |
azonenberg | you should never attempt to fit that much ram onto one address space | 23:33 |
azonenberg | Use hundreds of TB of ram by all means | 23:33 |
azonenberg | But not SMP | 23:33 |
azonenberg | use explicit NUMA aka message passing | 23:33 |
azonenberg | and keep your data local | 23:34 |
azonenberg | preferably in L1 cache | 23:34 |
berndj | hehe, one of my pie in the sky dreams is a attaxx machine that has TB of ram | 23:34 |
azonenberg | lol | 23:34 |
azonenberg | Mine is a little more reasonable | 23:35 |
azonenberg | only $6K and change | 23:35 |
berndj | it would basically be just a counter, some combinatorial logic, and huge continents of ram | 23:35 |
azonenberg | http://secure.newegg.com/WishList/PublicWishDetail.aspx?WishListNumber=12486029 | 23:35 |
berndj | because r is for random | 23:35 |
azonenberg | I did that workup a while ago just to see what it'd cost me to build lol | 23:36 |
azonenberg | maybe in five or ten years... :p | 23:36 |
berndj | i'm really sad that x86 has caused the demise of almost everything else | 23:37 |
azonenberg | I like mips a lot | 23:37 |
berndj | dec's alpha was ahead of its time, for example | 23:37 |
berndj | [where] does mips still get used, other than in undergrad CS courses? | 23:38 |
azonenberg | i didnt say it was popular :p | 23:38 |
azonenberg | But Its all over the place in embedded network hardware | 23:38 |
azonenberg | routing, switching, set-top boxes | 23:38 |
azonenberg | dsl/cable modems | 23:38 |
azonenberg | the majority of those are SoCs based on mips cpus plus external hardware | 23:38 |
berndj | oh yes. i remember cisco use[d] them | 23:39 |
azonenberg | pic32 is the only end-user marketed chip with a mips onboard afaik | 23:39 |
azonenberg | i tried to just buy like a raw mips cpu | 23:39 |
azonenberg | couldnt find any | 23:39 |
azonenberg | and pic32 doesnt have external memory buses | 23:39 |
azonenberg | so i said, screw that | 23:39 |
azonenberg | i'm building my own | 23:39 |
berndj | lol | 23:39 |
azonenberg | as soon as i work out this pipeline hazard and add a DRAM controller (and move to spartan-6 as the -3a i'm using is kinda cramped) | 23:39 |
azonenberg | i will have something pretty nice | 23:39 |
azonenberg | i plan to go multicore | 23:40 |
azonenberg | dual ~80 mhz 32-bit mips, maybe 64MB of sdram | 23:40 |
azonenberg | make an uber-low-power netbook | 23:40 |
azonenberg | we're talking something that is cold to the touch when in operation | 23:40 |
azonenberg | all passively cooled, no fans | 23:40 |
azonenberg | just a nice metal case that doubles as a heatsink | 23:40 |
azonenberg | throw in two 5000mAh li-po packs and you're good for 24 hours of operation | 23:41 |
berndj | PV keyboard | 23:41 |
azonenberg | The keyboard will be a dell netbook keyboard i found on ebay, most likley | 23:42 |
azonenberg | if and when i get the time to reverse engineer the matrix pinout | 23:42 |
berndj | did you build your first LED flasher when you were, like, 2 years old? | 23:43 |
berndj | i wish i had your focus / work ethic | 23:43 |
azonenberg | berndj: no, i started that when i was around 10 | 23:44 |
azonenberg | started coding in c when i was 9 but didnt get serious until 11 | 23:44 |
berndj | lol, only off by 7dB | 23:44 |
azonenberg | for the ifrst year or two it was just fooling around here and there | 23:44 |
azonenberg | didnt put a lot of time into it | 23:44 |
azonenberg | And lol, a lot of people say that | 23:44 |
azonenberg | There are costs to it, though | 23:44 |
berndj | i know :( | 23:44 |
azonenberg | for example, people expect you to be perfect | 23:45 |
azonenberg | they have higher standards than for anyone else | 23:45 |
berndj | i started programming only slightly older than you did | 23:45 |
berndj | oh, i was reading up on DNQ, and it looks like it might be kitchen-accessible to make if you're crazy enough to try | 23:47 |
berndj | start with mothballs | 23:47 |
berndj | but i'm still hoping for something less hairy | 23:48 |
berndj | around 250nm lots of possibilities open up, but another critical one closes: glass-based optics | 23:48 |
--- Sun Sep 18 2011 | 00:00 |
Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!