| 15:18 | mhoney | has there been a genral release date for the new cubox's? |
| 15:25 | dbsx | does anyone know what "UHS-I / UHS-1 (Ultra High Speed 1), Class 1 controller" means for an SDXC card? |
| 15:28 | jnettlet | dbsx, is means the Cubox-i supports the 1.8v signalling to support the higher data speeds for the UHS-I spec |
| 15:35 | dbsx | I got the quote from an SD supplier. So these cards (100MB/sread 45MB/s write) would work in a cubox-i? |
| 15:37 | cbxbiker61 | uhs-1 is downward compatible with non-uhs-1 controllers |
| 15:38 | cbxbiker61 | so the speeds are capped with non-uhs-1 controllers |
| 15:44 | dbsx | Do you know what sort of SD interface is in cubox and cubox-i? |
| 15:46 | cbxbiker61 | the old cubox is definately not uhs-1, not sure about cubox-i, but my guess would be non-uhs1....it seems like hardware vendors are not too excited about doing uhs1, i'm not sure why that is |
| 15:49 | dbsx | ok. I can live 20MB/s write (which what I get on my cubox) |
| 15:56 | jnettlet | dbsx, are you using them just for storage, but also writing small files? |
| 15:57 | jnettlet | That is what I was testing last week, a lot of cards get good write performance on large file sizes but choke on 4k writes |
| 16:06 | dbsx | I mostly get even performance. Small writes are lightning quick. (I must go back and check the numbers) |
| 16:07 | jnettlet | I have been using `iozone -e -I -a -s 50M -r 4k -r 512k -r 16M -i 0 -i 1 -i 2` for my testing. |
| 16:09 | dbsx | I will do that now. Give me a minute |
| 16:09 | jnettlet | no hurray :-P |
| 16:09 | jnettlet | hurray? |
| 16:09 | jnettlet | hurry |
| 16:09 | jnettlet | damn Danish is ruining my English |
| 16:19 | dbsx | jnettlet: what do you get on your cubox? |
| 16:21 | jnettlet | dbsx, depends on the card. |
| 16:21 | dbsx | on your good one |
| 16:22 | jnettlet | 1.9MB/s 4k random writes |
| 16:22 | jnettlet | That is a Sandisk Ultra 16GB HC-I u1 |
| 16:23 | jnettlet | They have to be the one that have 30MB/s in the title. |
| 16:23 | dbsx | I get 1.8MB/s on my USB hard drive |
| 16:23 | dbsx | the SD test is still running |
| 16:24 | jnettlet | some of SDHC cards are really really slow with 4k random writes, which is why running Android or Linux from them is so slow. |
| 16:26 | jnettlet | I had one card that wrote at .004MB/s, so 4KB/s. 1 write per second blech |
| 16:26 | dbsx | I currently have a 16GB SDHC (not an X). Waiting ... |
| 16:46 | jnettle | 16:46 * jnettlet has to run out for a bit. ttyl |
| 19:21 | jnettlet | rabeeh, how did testing go yesterday? Did you burnt he fuses? |
| 19:23 | rabeeh | yes. |
| 19:23 | rabeeh | :) |
| 19:23 | jnettlet | congrats |
| 19:23 | rabeeh | i have now 100 systems built; but we have a bit delay in shipping the samples |
| 19:24 | jnettlet | delays can always be expected with new hardware |
| 19:24 | rabeeh | we wanted to ship the samples tomorrow; but had some issues in the carrier board. |
| 19:24 | rabeeh | aargh |
| 19:24 | rabeeh | i hate delays; we will send probably with DHL to make it fast |
| 19:24 | rabeeh | we will ship Sunday instead |
| 19:25 | rabeeh | on that carrier board we don't want to put the original micro-sd; but the push-push (with the spring) |
| 19:25 | rabeeh | that's why there are delays. |
| 19:25 | rabeeh | jnettlet: i got your mail btw; thanks for the intro |
| 19:25 | rabeeh | i'll reply tonight |
| 19:26 | jnettlet | rabeeh, great. no rush. I have made sure everyone knows how busy you guys are at the moment. |
| 19:26 | jnettlet | oh so you are going with the push push, microsd? |
| 19:27 | rabeeh | for this to sample it out |
| 19:27 | rabeeh | to test the mechanics and the vendor |
| 19:28 | jnettlet | great idea. Which SOC did you use for this run? |
| 19:29 | rabeeh | the basic one |
| 19:29 | rabeeh | the solo / 512MB |
| 19:29 | rabeeh | but instead of fast ethernet we have gig |
| 19:29 | jnettlet | okay. |
| 19:29 | rabeeh | i know not the fanciest one |
| 19:30 | rabeeh | we will try to follow up with quads too |
| 19:30 | rabeeh | it's just a manner of changing the microSOM and keep exactly the same carrier. |
| 19:30 | jnettlet | after dealing with the MMP3, I completely understand why you would start with the the single core :-) |
| 19:31 | rabeeh | ? |
| 19:31 | rabeeh | from our side it's really inventory management |
| 19:31 | rabeeh | i think your story is totally different |
| 19:31 | rabeeh | mmp3 is ARM arch but prop processors --> lots of issues for vendors to debug |
| 19:31 | jnettlet | yeah, we are dealing with bugs bugs bugs |
| 19:31 | rabeeh | in freescale case it's a standard cortex a9 cpu complex |
| 19:32 | rabeeh | which rev is it running now? |
| 19:32 | jnettlet | how does the scaling work? Is it custom or a cpufreq module? |
| 19:33 | rabeeh | cpufreq |
| 19:33 | jnettlet | Our production run is b2 I think. We should have most the errata sorted out, but to keep our powersavings I really need Linaro's big.LITTLE work |
| 19:34 | jnettlet | Now that the 3.10 lts is out I am back playing around with getting it up and running. |
| 19:34 | rabeeh | does mmp3 have big little? |
| 19:35 | jnettlet | yep, 400mhz little and dual 1.2's |
| 19:35 | rabeeh | which processor is the little? |
| 19:35 | rabeeh | the codename? or the description? |
| 19:35 | rabeeh | is it a fixed stage pipeline? ARMv7-A instruction set? iwmmxt? neon? |
| 19:36 | jnettlet | armv7-a...processor is actually pj4b, it has iwmmxt and neon |
| 19:37 | jnettlet | trying to find my spec sheet. Not on my main computer right now. |
| 19:38 | rabeeh | yeah; i'm familiar with the pj4b; but that's the dual big |
| 19:39 | rabeeh | i was wondering about the little |
| 19:39 | jnettlet | that is also an armv7 and they call it a hybrid LPM (low power mode) core |
| 19:41 | rabeeh | now; why would the linaro big.LITTLE work do much for the OLPC? |
| 19:42 | rabeeh | i mean for a cell phone sending text message; or doing small things i do understand |
| 19:42 | rabeeh | but how handy would that be when using Sugar and all the stuff around it? |
| 19:43 | jnettlet | the dynamic plugging of the cpu's. So we can have each core completely off when not in use. |
| 19:43 | rabeeh | oh; so you can shut down one of the two big cores? |
| 19:43 | jnettlet | yep |
| 19:43 | rabeeh | or you mean the complex of the two bigs vs. the single little? |
| 19:44 | jnettlet | we can run just a single core and put the other two in wfe or wfi |
| 19:44 | rabeeh | ok; so you can't really power it off |
| 19:44 | jnettlet | not all the way, but we save about 300 mwatts per core |
| 19:45 | jnettlet | milliwatts |
| 19:47 | jnettlet | So being able to switch between running cores, lets us save even more power with our fast suspend mode. That is when there is no activity the DCON drives the lcd and we put the rest of the machine to sleep. Now when we get a wakeup event we can use the low power core to see what is going on before waking up the rest of the machine and possible using a faster core to do work. |
| 19:48 | rabeeh | what is the max power of the laptop today? |
| 19:48 | jnettlet | Machine is awake, we get a WOW event...wakeup little core, see if it is relevant, go right back to sleep. |
| 19:48 | rabeeh | i mean two processor + gpu + LCD on + wifi? |
| 19:49 | rabeeh | i know it's not an exact profile; but i'm looking at the ballpark to compare |
| 19:49 | jnettlet | I would have to double check, but the last time we tested that it was like 5.7 watts |
| 19:49 | rabeeh | backlight on max? |
| 19:49 | jnettlet | only the 2d core of the gpu...not 3d |
| 19:50 | rabeeh | oh; the 2d engine |
| 19:50 | jnettlet | yeah our screen only draws 1watt at max backlight. |
| 19:50 | jnettlet | My guess is 6.5 with 3d core |
| 19:50 | rabeeh | the 3d engine is a wattage black hole |
| 19:50 | rabeeh | i'v seen that on the Dove and i.MX6 |
| 19:50 | rabeeh | well... i'm not sure if it's the GPU itself or the DDR bandwidth that it creates when it works |
| 19:50 | jnettlet | yep, and as you know with the Vivante 1000/2000 there are two separate cores |
| 19:52 | jnettlet | yeah well the memory is something else that can be tackled. It is interleaved by default but does allow you to turn off backs. You would just need the kernel support to migrate memory first. |
| 19:52 | jnettlet | that would be something to add on later once kernel support is there. Much better for ereader mode |