#qi-hardware IRC log for Wednesday, 2014-06-18

nicksydneywpwrak: https://www.kickstarter.com/projects/1973729481/aurbee00:23
wpwrakmust be an interesting definition of "compatible' :)00:27
DocScrutinizer05x86 dma been slower than CPU sw mem-move since 80186 iirc02:00
DocScrutinizer05unless CPU does other stuff while DMA completes, it's way more overhead to set up DMA than to do same stuff in software. Even when both in the end reach same transfer rate. And nowadays DMA often is a relic like A20 gate, with CPU copying 10 words in the time where DMA moves one byte02:02
wpwrakwell, x86 dma is kinda special :)02:35
nicksydneyi don't read German but after translating it sounds interesting http://www.mikrocontroller.net/topic/33187610:29
whitequarkoh, that spectrophotometer thing again10:37
whitequarkit's practically useless10:37
nicksydneywhy useless ?10:39
nicksydney$2Million dollar is not useless :)10:39
whitequarkwell, it's useful to the ones producing it, sure10:39
nicksydneytime will tell whether they deliver or not..if they do deliver and backers are happy than it's worth the $2Million10:40
nicksydneywpwrak: it's using BLE too :)10:40
whitequarknicksydney: they probably will. it doesn't matter. it's a very expensive toy that you can't really use for anything interesting except "omg an apple"10:41
larscI don't think you need your smartphone to tell an apple and a banna appart ;)10:41
larscapelete: try to find out why and where the DMA based transfers spend there time compared to the PIO method11:54
larscmaybe that will tell you why it takes longer11:54
pcercueiDMA is working now?11:55
larsc00:36 < apelete> larsc: boot time with mmc dma mode + debug logs = 120s -> http://paste.debian.net/105528/11:57
larsc00:36 < apelete> larsc: boot time with mmc dma mode = 70s -> http://paste.debian.net/105545/11:57
larsc00:36 < apelete> larsc boot time with mmc pio mode = 34s -> http://paste.debian.net/105548/11:57
larsc00:37 < whitequark> wow, pio faster than dma11:57
larsc00:37 < apelete> larsc: it takes twice as long to boot with dma enabled, shouldn't dma be faster than pio ?11:57
pcercueiapelete: good job!12:07
apeletepcercuei: thanks :)12:50
apeletelarsc: I was thinking about using something like bootchart, but I guess it won't be detailed enough to spot dma transfers12:51
apeletelarsc: do you think ftrace could help ?12:53
pcercueiapelete: what CPU usage do you have during transfers? Ingenic's MMC driver for jz4770 uses DMA, is slow, and uses a lot the CPU12:54
apeletepcercuei: don't know yet, I only looked at the boot log timing12:55
apeleteI'm going to try a "zcat file >/dev/null" with PIO vs. DMA like wpwrak was suggesting last night12:56
larscapelete: Use something like getrawmonotonic() in the driver and messure the time it takes to transfer one block13:06
Action: DocScrutinizer05 wonders if they really got a proper mass spectrometer in that SCiO thing. If they do (I seen an early prototype of a sugar cube sized mass spectro several years ago at embedded fair in Nuernberg) then this could become a damn lot more than just "tell apat apples from banana" - it would be Abby's wet dream: Major Mass-Spectrometer in your pocket13:45
whitequarkDocScrutinizer05: it's NOT a mass spectrometer13:47
whitequarkyou can't have a mass spectrometer without at least a turbomolecular pump, and even if you have, it's more or less useless without a chromatograph13:48
whitequarkthey quite plainly explain that what they have is a simple spectrophotometer13:48
DocScrutinizer05the sugar-cube-mass-spectro was a damn smart ting: thanks to micro size they didn't need to evacuate the "chamber" where molecules get "sorted" in an electric field. And also thanks to the tiny size of the thing, really low voltages were sufficient to create the field to accelerate and deviate the molecules13:48
whitequarkDocScrutinizer05: your sugar-cube thing is NOT a mass-spectrometer13:49
DocScrutinizer05oooh, spectrophot is boring useless13:49
whitequarkit's also a NIR spectrophotometer13:49
DocScrutinizer05*my* sugar cube thing is https://en.wikipedia.org/wiki/File:Mass_Spectrometer_Schematic.svg13:50
DocScrutinizer05BZZZZ you're out13:50
DocScrutinizer05you may have a few buzzwords, maybe you find a link by yourself: mass spectrometer; fraunhofer; embedded Nuernberg13:51
DocScrutinizer05and a time frame: maybe 6 to 12 years ago13:52
whitequarkwas the first thing I did13:52
whitequarkno relevant results13:52
whitequarklink above is the closest one I found13:52
whitequarkthere's some interesting results from DARPA, but what they have is supposed to work on Mars13:53
whitequarkwhich means you don't even need to have a rough pump. considerably simpler13:53
DocScrutinizer05I talked with the guy and asked him "but how do you evacuate the thing?". He answered "we don't need to, the gap is so small only a few molecules come in anyway, no matter if it's air or the test substance"13:53
whitequarkdid he publish the result in a journal or something?13:54
whitequark(that'd make it easier to find)13:54
whitequarksaid DARPA thing, btw: http://www.gizmag.com/darpa-mems-smallest-vacuum-pumps/27883/13:57
DocScrutinizer05I asked him "but you need high voltages to accelerate and do the deviating field, no?"  he answered: " look at this tiny cube. Now figure we have hundreds of electrodes in there, each powered from normal 5V digital logic. We create fields of kV to MV per meter, thanks to the tiny structures"14:02
whitequarksince it's not evacuated, wouldn't such high potential gradient just cause a discharge?14:04
whitequarkand even if it would be14:04
DocScrutinizer05the gap is so small, there are only a few "dozen" molecules of anything in there at all14:04
DocScrutinizer05the space is too narrow to contain anything other than a pretty good vacuum14:05
DocScrutinizer05check out the gradients you have in CPU chips14:06
DocScrutinizer05when you power a modern chip with 5V instead of 1.2V, it probably has sparks all over the complete die ;-)14:07
DocScrutinizer05a trace with high potential and a ground level trace my be separated only a few dozen nanometers14:08
whitequarkhm, you're right14:08
DocScrutinizer051V difference makes for a gradient of 10E6 to 10E8V/m14:09
--- Thu Jun 19 201400:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!