| 08:07 | jnettlet[m]> | Just catching up on the chatter. Pretty exhausted after a good conference. We announced a lot of products with more to come. |
| 08:08 | jnettlet[m]> | Our first run of soms and boards have been brought up and memtested. I should have a running board by the end of the week and can start answering real world performance questions. |
| 08:08 | jnettlet[m]> | If there are specific tests / numbers anybody wants ping me and I will add it to my test suite. |
| 09:03 | jnettlet[m]> | FYI the iMX8M reference manual has already been released. |
| 09:17 | Ke> | yay |
| 09:18 | Ke> | btw. do you have any insight, whether puri.sm is able to get production quantities of i.MX8 M in time for librem-5 |
| 09:18 | Ke> | https://puri.sm/wp-content/uploads/2017/06/purism-librem-5-phone-planner.png |
| 09:19 | Ke> | In Q1 of 2018 we can get i.MX 8M into production. This is well ahead of our required hardware selection date of April 2018, so we will very likely be using the i.MX 8M in the Librem 5. |
| 09:24 | jnettlet[m]> | Ke: I doubt they were an early access partner like us and we still have a wait before we can get our devices into mass production. |
| 09:25 | jnettlet[m]> | Unless they worked out a deal with another early access partner. The chip has just gone into mass production. |
| 09:27 | Ke> | 2018 Q3 would apparently be first mass produced units |
| 09:27 | Ke> | or late Q2 |
| 09:28 | Ke> | We will be able to evaluate a i.MX 8M pre-production board November 2017 |
| 09:28 | jnettlet[m]> | I would guess Q3 may be a safe bet |
| 09:29 | Ke> | I guess I should pressure them for it, using A9 with spectre and meltdown fixes is not going to be fun to run a obsolete arch in 2019 |
| 09:30 | Ke> | in contrast with a53 which gets about the same perf with less power and no side channels |
| 09:31 | jnettlet[m]> | A9 doesn't need meltdown fixes. The current spectre patches are also pretty lightweight. |
| 09:32 | Ke> | ok |
| 09:32 | Ke> | though complete mitigation is a hard problem |
| 09:33 | Ke> | including jitted code |
| 09:33 | Ke> | especiall browser js |
| 09:33 | jnettlet[m]> | currently the mx8m is pretty close to the mx6 regarding power consumption. You just get additional graphics capability and a slightly higher clockspeed. But the in-order execution is of course a bit of a bottle-neck. The Cortex-a55 has a redesigned memory controller that helps to over come the bottlenecks of in-order execution. |
| 09:34 | Ke> | yeah I was drooling a bit over a55 |
| 09:34 | Ke> | there are not many SoCs that ship with it yet |
| 09:34 | jnettlet[m]> | I will start a developer blog for the iMX8M cubox and hummingboard. Help to keep everyone updated |
| 09:36 | jnettlet[m]> | We have also announced the new ClearFog....https://www.solid-run.com/product/clearfog-gt-8k-128emmc/ |
| 09:36 | jnettlet[m]> | definitely not a low cost developer board though :) |
| 09:37 | jnettlet[m]> | However I am tossing around an idea about providing some developer coupon codes for interesting projects. |
| 14:47 | topi`> | help, the hwclock on my Hummingboards are pointing at year 2005, how's it possible? |
| 14:47 | topi`> | it seems to save the sysclock just fine when using hwclock --systohc |
| 14:47 | topi`> | but after reboot, the clock is totally wrong |
| 14:48 | topi`> | hwclock --hctosys returns me to year 2005 |
| 14:52 | topi`> | sorry guys to bring you back to the past, the woes of i.mx6 :) |
| 14:52 | topi`> | setting the clock back to 2012 |
| 14:53 | topi`> | btw, regarding in-order vs out-of-order, I definitely prefer the better mem bandwidth on the cortex-a53, although you'll see some things like python interpreters be faster on the a9 cores |
| 14:53 | topi`> | a lower latency L2 cache would help those things as well |
| 14:54 | topi`> | I guess the a53 load/store units are radically better than on the a9 |
| 15:00 | Ke> | jnettlet[m]: does inc heatsink mean it's passively cooled? |
| 15:02 | jnettlet[m]> | Ke: we have a heatsink for it but it can only clock to 1.6ghz and needs to be mounted vertically to be passively cooled |
| 15:02 | topi`> | OK, I found out that there's two /dev/rtcs on HB, and the rtc1 is a SNVS which seems to work properly |
| 15:02 | topi`> | For the SNVS RTC, Freescale SNVS RTC support (CONFIG_RTC_DRV_SNVS), which is also enabled as built-in on the default kernel configuration file. |
| 15:03 | topi`> | jnettlet[m]: we need a turbo feature to run at 2.0 ghz until it's time to throttle down ;) |
| 15:04 | topi`> | an apt-get update can take lots of cpu time with fast network, that could be cut with a Turbo mode :) |
| 15:04 | topi`> | that's also why intel's chips "feel" faster |
| 15:04 | jnettlet[m]> | CPU frequency scaling is supported with Marvel's blob |
| 15:05 | topi`> | I'd love to see the A72 cores go as high as the intel/AMD cores, then we'd get some really nice performance |
| 15:06 | topi`> | I wonder what AMD does differnetly that their 28nm cores can easily reach 4.0 ghz |
| 15:06 | jnettlet[m]> | They aren't designed to clock that high |
| 15:06 | topi`> | the Bulldozer family cores definitely aren't any more performant than A72 |
| 15:07 | topi`> | I know, but why can Mediatek make A53 cores that go up to 2.2GHz even though the A53 was created for 1-1.5ghz? |
| 15:09 | jnettlet[m]> | Because it is a throw away soc. Meant to run for about a year or two |
| 15:10 | jnettlet[m]> | Like you said HummingBoards are meant to run forever |
| 15:12 | jnettlet[m]> | The SNVS rtc is only good for rtcwake. It doesn't have a battery backup so gets reset when it loses power |
| 15:16 | topi`> | does anyone know how to write a udev rule that would symlink /dev/rtc to /dev/rtc1? |
| 15:16 | topi`> | ah, ok, so no SNVS rtc... |
| 15:16 | topi`> | but, how can I get the rtc0 working? it always returns year 2005 |
| 15:16 | topi`> | rtc core: registered rtc-pcf8523 as rtc0 |
| 15:16 | topi`> | does this pcf8523 need some kernel patches? |
| 15:17 | topi`> | I'm running mainline 4.14 |
| 15:17 | topi`> | WTF, now it (rtc0) correctly returned 2018-03-02 after reset |
| 15:18 | topi`> | is this because I issued "hwclock -f /dev/rtc1 --systohc" just a minute ago? |
| 15:32 | topi`> | I'm creating a small module that allows programming the RTC after flashing, let's see if it pans out well |
| 15:32 | topi`> | but the damn buildroot doesn't actually have libc.so.6 at all! |
| 15:33 | topi`> | so my intent of using ctypes.CDLL("libc.so.6") works fine on my Debian PC, but fails miserably on buildroot's python |
| 15:33 | topi`> | (libc has the settimeofday() function which is your friend here) |
| 15:39 | topi`> | I still need to issue that hwclock though... |
| 16:58 | jnettlet[m]> | topi`: you just need to reset it and then then write the date to it. Or write the date to it twice |
| 17:29 | jas-hacks> | jnettlet have you started testing imx8m ? |
| 17:31 | jnettlet[m]> | jas-hacks: we just blue wired our first couple of dev boards. Now that EWC is over I should have my dev board within the week |
| 17:33 | jas-hacks> | it will be interesting to see what imx8m is capable of as it using the GC7000 lite |
| 17:35 | jas-hacks> | and whether the M4 can booted without waiting for the A53 cores |