| --- Day changed Tue Mar 31 2009 | ||
| kanzure | shape optimization with freefem++ - http://www.cmap.polytechnique.fr/~allaire/freefem_en.html | 00:04 |
|---|---|---|
| fenn | neat stuff | 00:14 |
| gene_ | http://www.technologyreview.com/computing/22339/?a=f | 00:15 |
| gene_ | check it out | 00:15 |
| gene_ | 500,000 of these chips should have around the same number of neurons as a human brain | 00:16 |
| fenn | blah.. i'd much rather that effort go towards an open source reconfigurable FPGA | 00:16 |
| gene_ | these people are brain researchers | 00:17 |
| gene_ | they aren't interested in such things | 00:17 |
| gene_ | but 500,000 is in the range of a production run | 00:17 |
| fenn | it's interesting that it runs 100000x faster than real neurons | 00:18 |
| gene_ | wonder if it'd become sentient | 00:18 |
| gene_ | damn | 00:18 |
| fenn | needs some i/o at the very least.. and real brains have a lot of hardwired functionality | 00:19 |
| gene_ | cameras | 00:20 |
| gene_ | the internet | 00:20 |
| gene_ | chat rooms | 00:20 |
| katsmeow-afk | i had some hope for the propellor chip, but it remains slow, and no way to tell which cpu is on the io buss at any time | 00:21 |
| katsmeow-afk | so the cpu are forced to share all external resources | 00:21 |
| fenn | propellor looks like an amateur first attempt at getting a chip fabbed | 00:22 |
| * katsmeow-afk has a stupid question: why not a state machine with one crosspoint set across them, instead of a box of neurons with the synapse possibilities fixed-in-hardware? | 00:23 | |
| katsmeow-afk | so for 50k neurons, you'd have at least 50k^2 possible connections? | 00:24 |
| katsmeow-afk | and that's thinking 2 dimensionally | 00:24 |
| katsmeow-afk | In fact, the current prototype can operate about 100,000 times faster than a real human brain. "We can simulate a day in a second," says Karlheinz. | 00:27 |
| katsmeow-afk | so if they multitask the thing, they can simulate 100k as much hardware with a system as slow as a human | 00:28 |
| fenn | yep | 00:28 |
| fenn | you need on-chip cache though | 00:29 |
| fenn | or else some really fast ram bus | 00:29 |
| katsmeow-afk | lay up some ram literally on top of each neuron, make a cube | 00:29 |
| fenn | i guess | 00:30 |
| fenn | how many layers can they do these days? | 00:30 |
| katsmeow-afk | dunno, but it bet it's not enough | 00:30 |
| kanzure | Markram and IBM were last working on some FPGA or hardwired ICs to simulate neurons | 00:31 |
| kanzure | what are these called again? AISC? fenn? | 00:31 |
| katsmeow-afk | however, if they use a serial data stream, recirculating, and all the ram sync'd, they could keep with the state machine idea and use only one wire to each neuron | 00:31 |
| fenn | ASIC = hardwired IC | 00:31 |
| katsmeow-afk | or 8,, or so | 00:31 |
| fenn | application specific IC | 00:31 |
| fenn | heh. bits in the pipe | 00:32 |
| katsmeow-afk | yeas | 00:32 |
| fenn | propagation delay ring buffer :P | 00:32 |
| katsmeow-afk | yeas | 00:32 |
| fenn | sort of gives a new meaning to volatile memory | 00:32 |
| katsmeow-afk | the neuron either changes the byte, or not, then it's put back in queue, and the next byte is sent in | 00:33 |
| katsmeow-afk | saves a huge buss into the chip, and an addy/data buss all thru it, and the muxes, 3-state buffers, etc | 00:33 |
| fenn | in brain research there are a lot more than 1 byte worth of variables | 00:37 |
| fenn | i'd hate to have to fab a new chip every time you want to add a variable | 00:37 |
| katsmeow-afk | so send in 2 or more | 00:38 |
| katsmeow-afk | well, atm, they have no option either way | 00:38 |
| fenn | ya | 00:38 |
| fenn | see why i'd rather have a reconfigurable FPGA | 00:38 |
| katsmeow-afk | but you get a ton less working density | 00:38 |
| kanzure | Eugen Leitl likes to complain about slow interconnection speeds | 00:39 |
| katsmeow-afk | cost may be even more that way, power use certainly would be | 00:39 |
| kanzure | for massively parallel computing. | 00:39 |
| fenn | the RC constant thing? | 00:39 |
| kanzure | ? | 00:39 |
| katsmeow-afk | i have some 600 128k 15ns ram chips,, to use a mass of them would cost so much more in power and pcb space, it's rediculus | 00:39 |
| fenn | long interconnects have high capacitance which slows down their rise times, introducing delay when communicating across the chip | 00:39 |
| fenn | so some smart kids tried making 'optical interconnects' basically fiber optics from one side of the chip to the other | 00:40 |
| fenn | katsmeow-afk: i dont know how the neuron simulator engine works on the chip, but i really doubt it's analog. so you'd be able to change the evaluation algorithm with an FPGA | 00:42 |
| fenn | um, i forgot where i was going with that | 00:42 |
| kanzure | was it something about IX10 or something? | 00:42 |
| katsmeow-afk | i forgot what it was even pertaining to | 00:42 |
| kanzure | I don't knwo what it's called, but there's an optical interconnect that supercloudcluster advocates like to rant about | 00:42 |
| kanzure | it has an 'X' in the name methinks | 00:42 |
| katsmeow-afk | kanzure, i remember that | 00:42 |
| katsmeow-afk | fenn, i never said cpld or fpga would not work, nor that they were analog | 00:43 |
| katsmeow-afk | i only said a mass of them might cost more than volunteer work at a fab place | 00:43 |
| katsmeow-afk | freeware, cost writeoff, volunteer, price paid in future stock, etc | 00:44 |
| fenn | i bet we will start to see spread spectrum chip-chip interconnects in a few years | 00:45 |
| fenn | instead of optical | 00:45 |
| kanzure | we don't have those yet? | 00:45 |
| katsmeow-afk | have you done the math on them? is one capable of handling 2.4ghz buss 64bits wide? | 00:45 |
| fenn | kanzure: i think they are stuck on just increasing clock speed with a plain squarewave signal | 00:47 |
| kanzure | http://www.maths.bris.ac.uk/~mairw/blog/archive/2006_09_10_archive.html | 00:51 |
| kanzure | anyone have experience with gerris? | 00:59 |
| kanzure | http://gfs.sourceforge.net/wiki/index.php/An_engineer's_pipe_flow | 01:11 |
| fenn | so you didnt know what to tell anythony lamantia? (any non-UT bio stuff in austin) | 01:30 |
| kanzure | I told him a few names | 01:30 |
| kanzure | a few groups, etc. | 01:30 |
| kanzure | just thought I'd forward his message to you and jonathan cline | 01:31 |
| katsmeow-afk | hmm Polycarbonate blocks all UVA and UVB rays | 02:08 |
| gene_ | I'm sure there's some non UT bio stuff in austin | 03:28 |
| gene_ | I see the come work at biotech company x poster in the halls every now and then | 03:30 |
| kanzure | gene_: did you see this? | 03:33 |
| kanzure | 03:33 | |
| kanzure | http://heybryan.org/bioreactor/membraneless_filtration/2009-03-30_2.png | 03:33 |
| gene_ | what the heck is that? | 03:35 |
| kanzure | just a navier strokes problem | 03:35 |
| kanzure | *stokes | 03:36 |
| kanzure | another: http://heybryan.org/bioreactor/membraneless_filtration/2009-03-30.png | 03:36 |
| gene_ | something has gone horribly wrong | 03:36 |
| kanzure | well the box is a little small, I'm going to increase it | 03:36 |
| gene_ | second one looks much better | 03:36 |
| kanzure | also it's not actually the problem that we want to solve, and the output is all wrong (who wants to look at vectors anyway?) | 03:36 |
| kanzure | ideally a colorful isosurface is what we should be looking at, or something | 03:37 |
| gene_ | wait a second are those vectors going through the wall? | 03:38 |
| kanzure | no | 03:38 |
| kanzure | they are just drawn like that :( | 03:38 |
| gene_ | umm | 03:39 |
| gene_ | it's in a box | 03:39 |
| kanzure | right, it's not on an infinite plane. | 03:39 |
| gene_ | yeah | 03:39 |
| gene_ | and that's an infinitely thin wall | 03:40 |
| kanzure | anyway, what I've been doing is going back through the literature and grabbing the right equations | 03:40 |
| kanzure | what's wrong with that? | 03:40 |
| kanzure | the walls are just boundaries, they are basically infinitely dense and ignored in the simulation | 03:40 |
| gene_ | document what equations you are using | 03:40 |
| kanzure | of course. | 03:40 |
| kanzure | http://science.slashdot.org/article.pl?sid=09/03/30/1218258 | 03:51 |
| kanzure | heh, go little turtle, go! | 03:51 |
| gene_ | do you know anything about curve fitting? | 04:02 |
| kanzure | functonal induction / regression analysis? | 04:02 |
| kanzure | *functional | 04:02 |
| gene_ | least squares method | 04:03 |
| gene_ | something like that I guess | 04:08 |
| kanzure | is there any particular way to figure out if a latex-math document is talking about the negative change-in-some-variable, versus the negative gradient vector? | 04:35 |
| kanzure | hm. context I guess. symbols without source code suck. | 04:37 |
| gene_ | http://en.wikipedia.org/wiki/Turboencabulator | 04:54 |
| fenn | holy crap it takes 45 minutes to fill up a standard 16GB class 6 SD card | 06:35 |
| willPow3r | sd cards aren't known for their high data transfer rate | 06:42 |
| fenn | yeah as i'm finding out | 06:44 |
| fenn | it might be worth spending 2x for 30MB/s data rate | 06:44 |
| fenn | vs 6MB/s | 06:44 |
| fenn | since i'll probably be all swapon swapoff like karate kid | 06:44 |
| willPow3r | it would cut the time to fill it down to 9 mins theoretically | 06:46 |
| fenn | yeah i'm actually not so concerned about that as much as swap speed | 06:46 |
| willPow3r | hdds have a much higher r/w speed than that | 06:49 |
| willPow3r | so why bother? | 06:49 |
| fenn | because i'll be strapping it to my leg | 06:50 |
| willPow3r | good reason. i figured you were using it in a standard computing environment | 06:50 |
| fenn | also thinking about getting multiple cards and doing RAID over USB :P | 06:51 |
| fenn | but i need one "main" card for the SD slot anyway | 06:52 |
| fenn | the adapters are $2 and a 4GB card is like $5 so why not | 06:53 |
| willPow3r | what are you doing with it anyway | 06:53 |
| fenn | i just need to find a small low-power USB2.0 hub | 06:54 |
| fenn | this is going to be my borg attachment versino 1.0 | 06:54 |
| fenn | "memory glasses" | 06:54 |
| fenn | i guess i should draw something up in inkscape | 06:54 |
| willPow3r | for augmented reality applications? | 07:08 |
| fenn | yes | 07:08 |
| fenn | and general not-tied-to-a-computer-ness | 07:08 |
| fenn | except i am, literally - that's the irony of it | 07:09 |
| willPow3r | it seems ironic but it really isn't | 07:13 |
| fenn | dealextreme has some weird stuff - Name Badge SD Card, Visor Eyeglasses Clip (type R for RACING), LED Light Illuminated Ear Pick | 07:13 |
| willPow3r | you're just getting rid of the ball-and-chain power cord | 07:14 |
| willPow3r | type R | 07:14 |
| fenn | http://www.dealextreme.com/details.dx/sku.6833 | 07:14 |
| fenn | it looks like a mutated USB key | 07:15 |
| willPow3r | it only got 3 stars out of 5 | 07:15 |
| fenn | well i can't figure out all these different lithium battery formats, so i'll just stick with AA NiMH | 07:16 |
| fenn | is there a name for this type of buckle/clip? http://www.amazon.com/Op-Tech-Connector-Uni-Loop-Customize/dp/B0010HA6A6/ref=pd_sim_p_3_img/192-9126761-4820535 | 07:51 |
| fenn | and the typical black nylon ribbon strap that it goes with | 07:51 |
| * fenn checks mcmaster-carr | 07:55 | |
| fenn | aha they are called webbing side-release or quick-release buckles | 08:08 |
| fenn | and i can even get it in kevlar if needed :D | 08:09 |
| fenn | not like the connector would survive that kind of treatment anyway | 08:09 |
| fenn | funny that the strap is so much stronger than the beefiest fitting | 08:11 |
| willPow3r | at least that way you know where the breaking point will be given enough stress to the system | 08:17 |
| fenn | apparently also called "fastex" buckles | 08:26 |
| fenn | hmm. today's lesson: make sure all the files are chowned to your username before uploading/extracting the .tar file | 09:14 |
| fenn | that is obnoxious | 09:15 |
| fenn | the wiki script makes new pages owned by www-data, and i just did chmod g-w somehow so i can't edit them | 09:17 |
| willPow3r | on flaminggay? | 09:17 |
| fenn | yeah | 09:17 |
| fenn | much obliged if you could chown fenn /var/www/fennetic.net/ -R | 09:17 |
| willPow3r | k done | 09:18 |
| fenn | ah. the bastard creates them as g-r | 09:20 |
| fenn | er, g-w | 09:20 |
| willPow3r | you want 2750 on the permissions? | 09:21 |
| willPow3r | they're set to 2770 now | 09:21 |
| willPow3r | so apache can write if you need to use a cms for anything | 09:21 |
| fenn | oh is that an apache setting? | 09:22 |
| willPow3r | apache runs under the www-data group | 09:22 |
| fenn | i want to be able to modify any files that are group www-data | 09:22 |
| willPow3r | if you use the # groups command you can see the groups that you are in | 09:23 |
| willPow3r | you should be in www-data as well | 09:23 |
| fenn | i am | 09:23 |
| fenn | i just made a new page in the wiki, and it created /var/www/fennetic.net/machines/wiki/6D6F72657374756666.txt | 09:24 |
| fenn | see the problem? | 09:24 |
| willPow3r | it created it as www-data:www-data | 09:25 |
| fenn | and -rw-r--r-- | 09:25 |
| willPow3r | yeah, i see it | 09:25 |
| fenn | which means i can't write to it | 09:25 |
| willPow3r | the stickybit isn't set for the directory. just a sec | 09:26 |
| fenn | i dont think that matters | 09:27 |
| willPow3r | try making a wiki page now | 09:30 |
| fenn | same thing | 09:31 |
| fenn | don't worry about it - i am dumping pukiwiki before long and i'll bug you then | 09:33 |
| willPow3r | heh k | 09:33 |
| fenn | or not.. i can just move the dir to a dark corner | 09:33 |
| willPow3r | still something i need to know how to fix | 09:33 |
| fenn | i am afraid it's some kind of php file creation mode setting | 09:33 |
| willPow3r | could be | 09:34 |
| willPow3r | its the umask setting in apache2.conf. i'll change it in a sec | 09:37 |
| willPow3r | its set, see if it works now | 09:43 |
| fenn | yes that worked | 09:44 |
| willPow3r | excellent | 09:45 |
| fenn | while you're at it.. the .htaccess files aren't doing anything | 09:46 |
| fenn | (allowoverride all) | 09:46 |
| willPow3r | done | 09:48 |
| fenn | cool thanks | 09:50 |
| ybit | kanzure, didn't you form an open manufacturing ieee online group? | 11:59 |
| ybit | ah, there it is | 11:59 |
| ybit | manufacturing automation | 11:59 |
| kanzure | you're going to run out of ip addresses | 13:06 |
| kanzure | does anyone know what Smari was talking about when he mentioned "permaculture"? | 14:05 |
| kanzure | I looked it over on wikipedia the other day and found that it started off like a reasonably good idea- taking responsibility for building human life support infrastructure- | 14:06 |
| kanzure | but then it turned into this terrible anti-tech campaign/movement thingy by the third or fourth sentence of the Wikipedia article | 14:06 |
| -!- any32072065 is now known as katsmeow-afk | 19:58 | |
| kanzure__ | I have a terrible idea | 20:04 |
| kanzure__ | compute the center of mass of the density distribution resulting from the flow through a channel of some arbitrary geometry | 20:06 |
| kanzure__ | compare the location of that center of mass to a target location (x,y) where you would prefer the density to end up at | 20:07 |
| kanzure__ | then use some sort of design algorithm (SA, GA, A*, BFS, DFS, ..) to 'design' (and then consequently evaluate) various geometries, and test them for that flow | 20:08 |
| kanzure__ | in particular I am thinking of test cases where an arbitrary channel geometry is specified but with one side as a straight line- so only one side is 'arbitrary' | 20:08 |
| kanzure__ | there are a few particular cases that I can imagine: (1) the line is defined as a sequence of different characteristics: rounded edges, jagged edges, etc., which are really just the perimeters of 2D shapes strung together; (2) specific cases where there are only a handful of variables to manipulate | 20:09 |
| kanzure__ | in case #2, it would be something like "vary the increaese in channel diameter by some function blah" | 20:09 |
| kanzure__ | or some other multivariable optimization problem. | 20:09 |
| kanzure__ | it was interesting that in my multivariable calculus class lecture today we were going over a density function for laminar objects and double integrals. haven't had such a ridiculously relevant lecture in a while now. | 20:10 |
| kanzure__ | woah wtf http://en.wikipedia.org/wiki/Mind_uploading flagged for deletion | 20:11 |
| kanzure__ | http://heybryan.org/books/papers/Cheap%20third-generation%20sequencing%20-%20nanopores%20-%20cyclodextrin%20-%20hemolysin%20-%202009.pdf | 20:16 |
| kanzure__ | Cheap third-generation sequencing via alpha-hemolysin nanopores and cyclodextrin. | 20:17 |
| kanzure__ | "an exonuclease digests single-stranded DNA, and as single bases fall into the pore, they transiently interact with the cyclodextrin and block an electric current that runs through the pore- the current amplitude is then easily converted into a known nucleotide" | 20:18 |
| kanzure__ | how sure are they that the nucleotide will fall into the pore? | 20:19 |
| kanzure__ | oh | 20:19 |
| kanzure__ | turns out they are struggling with that particular problem. | 20:19 |
| kanzure__ | oh | 20:31 |
| kanzure__ | so if you're able to discriminate between nucleotides based off of current, | 20:31 |
| kanzure__ | shouldn't the same be true for discriminating between two strands of DNA? | 20:31 |
| kanzure__ | you take a measurement of the current flow before the exonuclease chops off a nucleotide, and then a measurement after it chops off a nucleotide, and then you compare the two and if you have the ability to discriminate between individual nucleotides to begin with, then surely you have the same level of accuracy even with a much larger molecule? | 20:32 |
| kanzure__ | and the difference is then what's missing. | 20:33 |
| --- Log closed Tue Mar 31 22:11:04 2009 | ||
| --- Log opened Tue Mar 31 22:13:24 2009 | ||
| -!- Irssi: #hplusroadmap: Total of 15 nicks [0 ops, 0 halfops, 0 voices, 15 normal] | 22:13 | |
| -!- Irssi: Join to #hplusroadmap was synced in 92 secs | 22:14 | |
| kanzure | lesson learned: never ever call apt-get -t unstable r-base | 23:31 |
Generated by irclog2html.py 2.15.0.dev0 by Marius Gedminas - find it at mg.pov.lt!
