#qi-hardware IRC log for Sunday, 2011-08-21

wolfspraulthe 4760-based hardware we can get now has way too many closed parts embedded to be fun to work with00:05
wolfsprauland they were not put there by accident, but on purpose. makes no sense to me to work against that somehow, let them do what they want to do...00:05
wolfspraulI'm wondering what the mechanical of the Ya NanoNote will look like, and how it's made00:06
wolfspraulwe ended up doing nothing for the ben-wpan mechanical, and I think that's a little sad. as an open pcb they are very vulnerable.00:07
rohoh.. what parts are closed? some 3rd party gpu?00:08
kristianpaulSIMD :-)00:09
kristianpauli remenber...00:09
wolfspraulI meant the entire product. schematics, layout, bom, datasheets of all chips, mechanical, etc.00:11
kristianpauloh sure...00:11
kristianpaulatben case, i did some tests in the past actually, i still having the prototypes with me.. but a 3d printed case is not the best quality in the short term of use..00:15
kristianpaulwell there are other steps that can improve that but not my interest right now..00:15
wolfsprauloh sure, just saying which 'todo' parts I'm thinking about00:16
wolfspraulfinding the shortest way to a 4760 product is not my thinking. if it's easy somewhere and helps, we take it.00:16
rohtrue. there is a lot to optimize. but anyhow, we need a working, not too expensive soc which is faster and can do more than the old one in the ben00:17
rohatleast thats my pov00:17
wolfspraulall just opinions anyway. when designing a product, the strength and flexibility of free software can overcome the advantage of a new piece of closed hardware, imho. at least sometimes, and I am looking for those kinds of opportunities.00:17
wolfspraulkristianpaul: any namuru / tracking news?00:18
kristianpaulnot yet00:18
kristianpauli have some to do actually...00:18
rohso we can concentrate on stuff like mechanics and not reinvent the soc also in the next product00:18
kristianpaulit seems so..00:19
wolfspraulroh: thinking about anti-vendor just now, look how the Palm Pre ended up00:25
wolfspraulall that passion was wasted again00:26
rohi dont get the anti-vendor part00:26
wolfspraulwe need to get smarter in how to build a lasting foundation00:26
rohi mean.. would you rather like to use something like a freescale cpu?00:26
wolfspraulno I just meant in general, same as your references to '4760 products in the open market'00:27
wolfspraulif there are vendors we buy from, they should support the openess00:27
wolfspraulemail the vendor who makes that 4760 hardware, ask them to publish or at least send you schematics, layout, bom, all datasheets, mechanical design files, etc. if they don't reply, don't use that stuff :-)00:28
wolfspraulin the few cases I tried since (like Hanvon e-readers), I failed, predictably. Waste of time.00:28
wolfspraulso I'm trying to find more companies and partners that share our vision on the power of open parts.00:29
wolfspraulwhere I believe we make nice progress00:29
wolfspraulroh: how do you like the Freescale chips? anything that stands out, is interesting?00:34
wolfsprauljust reading about Freescale on Wikipedia - wow they have interesting stuff!00:37
wolfsprauland what a history. first two-way mobile radio etc.00:38
rohwolfspraul: well.. the chips seem to have solid linux support done by people who know what they are doing00:48
rohspecs and datasheets are open to anyone without registration..  from my pov pretty much as good as it gets00:49
rohespecially since support is mainline already mostly afaik00:49
rohpengutronix is doing the kernel stuff as far as i know00:49
wolfspraulthe only problem is the price00:53
rohyou think?00:53
wolfsprauljust checking at digikey a bit. they have about 20 Freescale i.mx in stock.00:53
wolfspraulthe cheapest costs 9 USD, i.mx 2300:54
wolfspraulthen it quickly goes up to 20, 30, more00:55
wolfspraulthe cheapst i.mx51 costs 38 USD00:55
rohsure. checked the specs?00:55
wolfspraulthe cheapest i.mx 50 23 USD00:55
wolfsprauloh I can imagine the specs. but those prices pretty much destroy any hope to make powerful cheap devices, ever.00:56
rohalso... people are happy to pay 20-30% of the product price (compared to single unit prices) on desktop and notebook cpus. so why should that be a problem for other devices?00:56
wolfspraulbecause the new chips will be priced as high or even higher, and the old ones discontinued to not cannibalize sales of the new chips too much00:56
wolfspraulwait wait00:57
wolfspraulthere's so many things that go in there00:57
wolfspraulthere is _zero_, but really _zero_ innovation in the desktop and notebook space00:57
wolfspraulthose companies have collectively let go anybody who could innovate00:57
wolfspraulso yes, the percentage of the cost of the main CPU in a typical notebook is amazing00:57
rohimx 50 is 800mhz and ddr2 ram00:58
wolfspraulthe i.mx chips would only be the central part of any interesting product00:58
rohthe slowest imx is ee has 400mhz00:58
rohhttp://www.freescale.com/files/32bit/doc/brochure/FLYRIMXPRDCMPR.pdf seems a good overview00:58
wolfspraulyes but look at the price structure. 10-40 usd basically. that price structure will be kept, it may even go up.00:58
wolfspraulthat's what gets you the nice kernel and pegutronix and all :-)00:59
wolfspraulI'm not in favor or opposed or anything, just trying to get an overview over data points.00:59
wolfspraulbut that kind of pricing for 'just' the cpu of your device will, like I said, destroy any hope of making cheap mass-market thingies00:59
wolfspraulit works for a lot of more specialized vertical markets though01:00
rohsure. i am just thinking loudly how we can reduce work on nasty stuff and focus on stuff we can 'gain' some new knowledge and pull it into the open.01:00
wolfspraulthe good stuff I see now is kicad, schhist/brdhist, boom, dsv, openerp (still learning)01:01
rohand if we need time and energy to focus on process and mechanics, i guess its ok to 'outsource' some of the time consuming development work of doing platform support to somebody else trustworthy.. i guess pengutronix and mainline is as good as that can get right know01:01
wolfspraulnot switching to a freescale i.mx (I just speak for myself of course)01:01
wolfspraulbut the Ben NanoNote works, it's already designed, produced, etc.01:01
wolfsprauland I will go back to Ingenic, just needs the right reason.01:02
rohtrue. still keeping the soc is a gurantee to fail in my eyes.01:02
wolfspraulFreescale and Ingenic are not comparable I think, 2 totally different companies.01:02
rohgoing to the next soc from them is not so much less work than going to a working platform with linux port.01:02
wolfspraulI think the kernel we have on the Ben is outstanding.01:03
rohfrom my pov Ingenic only HAS a working linux port because of the ben.01:03
rohso all nasty, time consuming work has been done here. and that will happen again with a new soc where that has not happend before (new ingenic soc)01:04
wolfspraulwait wait01:05
wolfspraulfirst there is quite a bit of 4760 support already01:05
wolfspraulsecond - I am not actively working on a 4760 product now, so I don't have that problem :-)01:05
wolfspraulthird - I can go back to Ingenic at any time and demand things from them, whether they do them or not is another question01:05
rohsure. the whole thinking only comes into place then anyhow.01:05
rohmy point is: neither freescale NOR ingenic are capable of doing a proper linux port.01:06
rohelse it would be there already.01:06
rohspecialist companies who know how linux works do such things. for > 1 vendor.01:06
wolfspraulI described the low hanging fruits to you above. kicad and tool family, fped, boom, openerp, the entire mechanical toolchain and process01:06
rohvendor ports are always _crap_01:06
wolfspraulwe have to get the cost of small runs of interesting products down01:06
wolfspraula small run of a qfp chip on a breakout board is always easy01:06
wolfspraulso that's not what I mean01:07
wolfspraulbut if we can effectively do small runs of medium or higher complexity anything (like the m1), then we have a real chance in the market.01:07
rohall valid points. i just dont want to make the same stupid errors like on openmoko, where keeping a shitty, not fitting chipset together with a overused base design made things stall way to much and delayed everything.01:08
wolfsprauland then we can pick the hardware suppliers that help us make our products work, which can include Ingenic, or Freescale, or others.01:08
wolfspraulif we could time travel back there we would probably do everything differently01:08
wolfspraulit would start with not adding all those new chips on gta02 ;-)01:08
wolfsprauland moving gta01 production to taiwan. and spending much more time on the process, sourcing, tools, production testing, etc.01:09
rohi think mm1 is already our current limit of complexity. we see that on all kinds of things.. layout problems (emv, etc) manufacturing precision, sourcing...01:09
wolfspraulgetting the yield under control and getting knowledge of yield problems to more people.01:09
wolfspraulanyway, hindsight thinking01:09
wolfspraulyes, mm1 is nice01:09
wolfsprauljust the right level. we can grow without dying :-)01:10
wolfspraulnear death maybe...01:10
rohin hindsight.. gta01 should never have been sold. and gta2 been developed learning and revising stuff , not adding complexity without testing all things first (prototypes and real tests)01:10
wolfspraulgta01 was a run of 3000, it has to be sold01:11
wolfspraulor a run of that size was premature01:11
rohi know. much too many01:11
wolfspraulyou can accept the first 'few hundred' as part of the ramp-up process, but not more than that01:11
wolfsprauleven a few hundred only works for cheap consumer electronics01:11
wolfspraulotherwise you need to slow down the ramp-up and learn more first. no point in producing more and more electronic garbage.01:12
rohi think we need to find new levels and process steps for the pvt and dvt series01:12
rohif we go too low, we will not find bugs (sample size too low). if we go too high, we waste money.01:12
wolfspraulcorrect :-)01:13
rohbut i think we are on a good way there.01:13
wolfspraulI'm wondering how some of those ARM chipmakers continue01:23
wolfspraulI wouldn't want to be an ARM chip designer. Like TI for example - will they sell OMAP? To whom?01:23
rohthey do sell.01:24
wolfspraulMarvell will become a company just designing chips for their largest customers, they will get out of the general arm chip designer market.01:24
rohi mean.. just watch the asian market and sales for android devices01:24
wolfspraulwhat chips are in them? I cannot imagine many OMAP designs.01:24
rohall that iOS vs android stats are basically bullshit because they neglect the inner-asian markets completely01:24
rohe.g. there are ONLY 3 vendors selling android3 devices now. samsung, toshiba and acer. and all are nvidia tegra201:25
wolfspraulhow does nvidia design and sell their chips? don't know either. argh. the market is so big.01:25
rohthe only one not using tegra announced on omap4 and is some company from china? ive never heard from01:25
wolfspraulyes nvidia must be doing something right.01:25
rohwell.. yes and no. nvidia is closed source. period. will not change (they cant, too much licensed 3rd party afaik). i dont buy stuff from them.01:26
kristianpaulwhat about motorola, they ship android 3 too01:26
rohthey do?01:27
rohwell.. they got bought by google, didnt they?01:27
wolfspraulI cannot find nvidia tegra on digikey or octopart.com at all01:27
rohmostly for patents afaik, but we will know soon enough if they will continue01:27
rohwolfspraul: heh. no chance.01:27
kristianpaulbbl :)01:28
wolfspraulok so most likely they do designs in close collaboriation with their top-3/4 customers then01:28
rohwolfspraul: they are even too closed to name a price. nvidia has nothing to do with a 'open market'01:28
wolfspraulwell I understand that.01:28
wolfspraulyou may simply misunderstand their business model01:28
wolfspraulif you go to Apple to buy (eatable) apples, they will also turn you away friendly :-)01:28
rohack. thats why i named freescale. their businesscase is long term hw like embedded and industrial stuff. much more diverse than the '10-20 devices/vendors' max we see from nvidia or so01:29
wolfspraullook at freescale's entire range (I just read their wikipedia page)01:29
rohso they are used to small customers01:29
wolfspraulthen you will see that digital cpus are just a small, very small, part of their business01:29
rohatleast for the imx stuff01:29
wolfspraulwhich is not bad, that's how they fit in there01:30
wolfspraulthey need them so they can have a full range of chips for their customers01:30
rohack. and i trust them not to flip around policy too often since it would hurt their businesscase and confuse their long term customers.01:30
wolfspraulso yes, I agree, it looks like Freescale will continue to design general market arm chips01:30
wolfspraulsure, they cannot. look at their other chips.01:31
wolfspraulit's almost as if Analog Devices would start making some arm designs, to better integrate with their data converters (they may even do just, I'm just picking an example out of the air)01:31
wolfspraulwith TI I'm not so sure, I can imagine them selling OMAP01:32
rohack. or flip policy around next chip and closing everything up again01:33
wolfspraulthey are too big to do what Marvell and nvidia are doing. and taking on Intel directly with "the fastest arm in the world" is too risky.01:33
wolfspraulwho else is left in arm land?01:34
wolfspraul#1 freescale, we agree01:34
wolfspraulnvidia, marvell, samsung - no01:34
wolfspraulti - not sure, don't feel comfortable on their strategy01:34
rohbut different scale (small mcu, fighting for parts of the avr market)01:34
wolfspraulonly freescale, phew01:35
wolfsprauland then prices are high. you cannot build a competitive smartphone (for example) around a freescale i.mx01:35
rohif you ask for big names, yes. there are hundredts of small companies as ingenic i guess (and which i cannot name)01:36
wolfspraulhundreds maybe a bit too much01:36
wolfspraulin China I know 5-1001:36
rohwolfspraul: even if you would ADD the price of the cpu to a free product it wouldnt sell that worse.01:36
rohstop fighting for a cheaper devices. make a better one. thats the only way to suceed.01:37
wolfspraulyes and no. price is an important part of 'better'.01:37
wolfspraulI am not fighting for cheaper devices, I am fighting for a cheaper process.01:37
wolfspraulbig difference01:37
rohit cannot not to be twice as expensive as the nonfree stuff, but it can have some decent (10-20% up-pricing) which people will happily pay for some more freedom01:38
wolfspraulthe process has a lot of potential for optimization01:38
wolfspraul10-20% up from a 30 USD Android smartphone? :-)01:38
wolfspraulenough for a coffee at Starbucks01:38
rohack. so don't frown on a cpu which is 10E instread of 5.. these 5E i guess every customer made happy will pay gladly (i am and would)01:38
wolfspraullet's get the process optimized with free tools, then we can make winning products. I hope mm1 is a start, and the Ben has potential too, in its current form or its successor.01:39
wolfspraulthe mechanical is as important as the cpu01:40
kyakmth: (on button fast) yeah, i know it is already addressed in 3.x kernel. But we need it now, in 2.6.37 :)07:01
mthkyak: the commit should be easy to port to an older kernel13:24
mththe only reason it was added in 3.0 and not before is that I wasn't aware of the issue before13:24
mthI think it was zear who told me about it13:24
zearmth, i told you about what now? :)13:26
mththat the Dingoo took a lot of time to power on13:28
kyakmth: do you see any consequences of doing it at boot time via poke?13:28
mthas long as the registers are initialized some time it shoudl work, I think13:29
kyakbecause this is just faster than porting back even trivial patches.. And already working :)13:29
mthah, if it's already working then it's fine13:30
mthI didn't read the entire discussion; I thought you might still be looking for a solution13:30
kyakyep, it does work like this13:30
kyakmth: btw, do you have an intention to add screen-blank functionality to gmenu2x?13:32
kyakmost of the time, i just go to console from gmenu2x so the screen would go blank13:32
mthit's already present, but there is a problem with SDL where a thread still exits after exec-ing the selected application13:53
mthsee Settings > Screen Timeout13:54
mthit was added a few months ago13:54
mthAwAyla has been looking for the cause of the extra thread and did get some clues, but hasn't found it yet13:54
mth*exists, not exits13:55
kyakoh, it's been set to 013:58
kyaklet's see if it works now13:58
kyakall right, it's working from gmenu2x itself!13:58
kyaknow, should i start some GUI application form gmenu2x and see if screen goes blank?13:59
mthit only applies to gmenu2x itself14:01
kyakso it works fine14:24
kyakhm, now i wonder where is it controlled for console.. setterm -blank 30 doesn't seems to affect it14:31
kyakhah, it's minutes, not seconds :)14:33
qi-botThe build has FAILED, see log here: http://fidelio.qi-hardware.com/~xiangfu/compile-log/openwrt-xburst.full_system-08212011-0239/14:34
kyakseems that mocp fails to build with updated ffmpeg14:37
kyakaccording to google, the next failure will be mediatomb14:55
kyakjow_laptop: do you think jumping from ffmpeg 0.5.4 to 0.8.2 was a good idea? At least this should have been tested first against dependant packages...14:58
jow_laptopkyak: no I don't think so14:59
kyakffmpeg has deprecated an older API, so expect a lot of breakage..14:59
jow_laptopI mean I don't think it was a good idea but honestly I only care for backfire atm14:59
jow_laptopwhich has a forked packageset15:00
jow_laptopbut I'll talk to the committer15:00
kyaka forked packageset? this is interesting15:00
jow_laptopwell I froze packages/ beginning with rc515:01
kyakbecause it was always a fight to keep packages compilable for backfire :)15:01
jow_laptopI now only backport fixes15:01
jow_laptopit will remain this way until final15:01
kyakcool. I didn't think there going to be a backfore-based release15:01
jow_laptopapparently it has bad karma .)15:01
jow_laptopevery release manager so far disappeared sooner or later15:02
jow_laptopnow I overtook and promplty got swamped by paid work :P15:02
kyakare you the release manager?15:02
kyakheh :)15:02
kyakmaybe being a release manager for backfire is a good karma only for the release manager himself? :)15:03
wpwrak"poke" is part of the boot process ? now i really have to catch up with that thread ;-)15:31
kyakwpwrak: did you catch up?17:17
wpwrak_kyak: just with IRC so far ;)19:23
ignatius-Is it possible to "fsck" the Ben remotely? I have some UBI errors that need to be fixed, which don't allow me to boot into the system.20:17
larscyou could boot from mmc20:43
ignatius-I tried. Didn't work. :( F1/F4/U/M/S)20:45
ignatius-I think I may have an older version of the boot loader.20:45
ignatius-Someone in here said there was a "bad" version of it. I tried compiling an earlier version of the XBurst tools... same problems.20:46
ignatius-As well as the newest branch.20:46
ignatius-Any idea where I can get the newest u-boot image?20:51
ignatius-Same version I already had. :(20:56
ignatius-Ok... what about the newest XBurst tools?20:56
ignatius-I can't figure out why F1+PowerON and F4-PowerON doesn't work.20:58
ignatius-M+PowerON doesn't.. S+PowerON doesn't..20:58
--- Mon Aug 22 201100:00

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