**** BEGIN LOGGING AT Tue Sep 24 02:59:59 2019
**** BEGIN LOGGING AT Tue Sep 24 09:57:36 2019
Sep 24 14:03:24 <kremlin> m
Sep 24 14:25:22 <KotH> m?
Sep 24 14:25:34 <KotH> "m" as in "mhmmm... chocolate!" ?
Sep 24 14:35:08 <kremlin> "good morning"
Sep 24 14:52:53 <zmatt> I like the "mhmmm... chocolate!" explanation more
Sep 24 14:57:17 <GenTooMan> chocolate isn't a bad thing really..
Sep 24 15:32:02 <Bob31> My BeagleBone AI kernel say critical temperature reached (80 C), shutting down
Sep 24 15:32:18 <kremlin> there's a high-temp BBB model iirc
Sep 24 15:32:29 <kremlin> automotive or military grade rated iirc
Sep 24 15:32:36 <zmatt> kremlin: that's not what he's talking about
Sep 24 15:32:42 <kremlin> oh?
Sep 24 15:32:59 <Bob31> This is after about 3 minutes on time
Sep 24 15:33:06 <zmatt> I've seen on the list that the BBAI indeed has thermal issues, especially with certain kernels where power management isn't working properly
Sep 24 15:33:12 <kremlin> ah
Sep 24 15:33:44 <zmatt> Bob31: your best bet is to just follow the mailing list / forum topic on this subject for suggestions and new developments
Sep 24 15:33:47 <kremlin> that makes sense -- 80 C is...hot
Sep 24 15:34:33 <zmatt> hmm, assuming it's public
Sep 24 15:34:38 <Bob31> I did upgrade as recconmended in the start guide
Sep 24 15:35:53 <Bob31> zmatt I did try to search for the prob
Sep 24 15:36:11 <zmatt> hmm maybe there's no discussion topic yet on the public mailing list I just realized
Sep 24 15:36:40 <zmatt> the BBAI has been released only a few days ago, so most of the discussion is stlil on the private pre-release list
Sep 24 15:37:12 <zmatt> maybe just start a topic about it here => https://groups.google.com/forum/#!categories/beagleboard/beaglebone-ai
Sep 24 15:44:06 <Bob31> zmatt Thanks, I will ask Q on forum
Sep 24 19:01:03 <zoltan> hi
Sep 24 19:10:26 <kzoltaan> hi, I received my new bb ai, and get this message. kernel:[ 2364.467158] thermal thermal_zone0: critical temperature reached (95 C), shutting down // only apt-get install runs...
Sep 24 19:11:35 <zmatt> jkridner[m]: was it really a good idea to release the BBAI before this problem was solved? -.-
Sep 24 19:15:41 <kzoltaan> very hot.
Sep 24 19:17:11 <jkridner[m]> seems to be worse than I expected. also, the wrong .dts got pulled into the on-board image in production.
Sep 24 19:17:26 <jkridner[m]> if you can get a good fan on the device and do the update, it should help a lot.
Sep 24 19:18:45 <zmatt> wait the AI actually includes a heatsink and it still overheats for people just running idle? my bbx15 doesn't have a heatsink and it's pretty warm to the touch when idle but I've never had a thermal shutdown, even during periods of high activity such as installing updates or compiling stuff
Sep 24 19:19:07 <zmatt> what's causing the difference?
Sep 24 19:21:11 <zmatt> thermal values on my bbx15 running idle: https://pastebin.com/raw/9XVYxGn1
Sep 24 19:22:11 <zmatt> (cpu governor "performance", both cores 1.5 GHz)
Sep 24 19:30:28 <kzoltaan> root@homecontrol:/sys/class/thermal# cat thermal_zone*/temp
Sep 24 19:30:29 <kzoltaan> 72200
Sep 24 19:30:29 <kzoltaan> 71400
Sep 24 19:30:29 <kzoltaan> 71000
Sep 24 19:30:29 <kzoltaan> 71400
Sep 24 19:30:29 <kzoltaan> 71400
Sep 24 19:30:40 <kzoltaan> sorry
Sep 24 19:30:43 <zmatt> please don't paste multiline output into chat
Sep 24 19:30:56 <kzoltaan> yes, im sorry.
Sep 24 19:31:06 <zmatt> np :)
Sep 24 19:31:55 <kzoltaan> but the my mono apps run very fast on this device.
Sep 24 19:32:21 <zmatt> echo $(cat thermal_zone*/@(type|temp) )    is a nice way to get the data as one-liner (requires the extglob shell option, i.e. "shopt -s extglob")
Sep 24 19:37:05 <kzoltaan> 76200 cpu_thermal 74100 gpu_thermal 75000 core_thermal 73500 dspeve_thermal 74600 iva_thermal
Sep 24 19:37:32 <jkridner[m]> watch /opt/scripts/device/x15/test_thermal.sh
Sep 24 19:41:19 <kzoltaan> 85400 cpu_thermal 84600 gpu_thermal 83800 core_thermal 84200 dspeve_thermal 84200 iva_thermal
Sep 24 19:43:03 <kzoltaan> 89000 cpu_thermal 86200 gpu_thermal 86200 core_thermal 86200 dspeve_thermal 87400 iva_thermal
Sep 24 19:45:53 <zmatt> I just quickly made this script to quickly monitor temperature => https://pastebin.com/Cj1hQuUr  dunno how it compares to the script jkridner[m] just mentioned (I don't seem to have it)
Sep 24 19:46:00 <zmatt> brb, need to go to the store
Sep 24 19:50:55 <zmatt> eh, too lazy... whatever I'll just order some food
Sep 24 19:53:41 <zmatt> just did an apt-get dist-upgrade: https://pastebin.com/raw/zmAQvT10
Sep 24 19:58:21 <zmatt> let's see what happens if I cause 100% cpu load on one core
Sep 24 19:58:35 <kzoltaan> https://pastebin.com/raw/RFqQ6XSr
Sep 24 19:58:47 <zmatt> 75
Sep 24 19:59:33 <zmatt> 76
Sep 24 20:00:33 <zmatt> it's definitely not rising enthausiastically, but it's still rising slowly, probably because the board is warming up (53)
Sep 24 20:00:49 <zmatt> 78
Sep 24 20:01:19 <zmatt> but this is for several minutes of 100% load on one core already
Sep 24 20:01:29 <zmatt> 79
Sep 24 20:02:51 <jkridner[m]> less copper might be part of the issue.
Sep 24 20:03:05 <jkridner[m]> getting a fan on it and updating the device tree should help.
Sep 24 20:03:11 <zmatt> but it has a heatsink, I don't
Sep 24 20:04:39 <jkridner[m]> https://beagleboard.org/upgrade/
Sep 24 20:05:03 <jkridner[m]> yeah, seems odd. sample size is definitely a lot larger now.
Sep 24 20:05:17 <zmatt> I've been running my bbx15 at a loadavg of 1.00 for more than 7 minutes with no heatsink or fan, and it's still having trouble reaching 80 degrees
Sep 24 20:05:26 <jkridner[m]> wouldn't exclude an issue with the .dts
Sep 24 20:06:57 <zmatt> I guess there's no way to determine how much current is being consumed on the various supplies, to see if maybe there's somehow a high I/O current draw?
Sep 24 20:07:50 <zmatt> it seems I've levelled off at 80 degrees cpu, 56 degrees board
Sep 24 20:10:03 <zmatt> I'm using kernel 4.14.62-ti-r69 fwiw
Sep 24 20:11:02 <kzoltaan> whats wrong in dts?
Sep 24 20:28:01 <kzoltaan> died again. 95C reached....
Sep 24 21:29:59 <jkridner[m]> rcn-ee: seems like ti-tidl is missing the .h files and other helpful parts.
Sep 24 21:32:47 <jkridner[m]> rcn-ee: https://github.com/beagleboard/cloud9-examples/issues/16
Sep 24 22:09:22 <rcn-ee[m]> fixing incoming: https://github.com/rcn-ee/repos/commit/107dfacb62567650c3b4173b17708742b1dd58f8
Sep 24 23:08:48 <ds2> hmmmmm
Sep 24 23:08:58 <ds2> how available are the BBAI's at this momemnt?
Sep 25 00:15:58 <rcn-ee[m]> ds2 instock
Sep 25 00:58:41 <ds2> rcn-ee[m]: is it remaining in stock though? trying to make sense of the heat complaints
Sep 25 01:13:39 <rcn-ee[m]> it should, for this week.. it gets warm, 'fan cape' should be available soon, a desk fan works fine..
Sep 25 01:18:23 <ds2> wouldn't a bigger heatsink be just as effective?
Sep 25 01:18:49 <ds2> something like a small heat pipe assembly to a bigger dissipiator so the 2 connectors aren't impacted
Sep 25 01:19:58 <rcn-ee[m]> ds2: i want to build a bracket to utilze this: https://www.seeedstudio.com/ICE-Tower-CPU-Cooling-Fan-for-Raspberry-pi-Support-Pi-4-p-4097.html
Sep 25 01:20:03 <zmatt> why does it get so hot though while my heatsinkless bbx15 levels off at 80 degrees under 100% cpu load on one core?  is my x15 just unusually chill?
Sep 25 01:20:23 <rcn-ee[m]> the x15 extracts more heat thru the pcb..
Sep 25 01:21:28 <ds2> rcn-ee[m]: but that would block the connectors, right?
Sep 25 01:22:15 <rcn-ee[m]> details. ;)
Sep 25 01:22:29 <zmatt> and that's more effective than a heatsink?  enough so to allow mine to almost remain cooler under 100% load than some people's BBAI seems to be when idle?
Sep 25 01:22:31 <rcn-ee[m]> just think it would look cool to have a massive heatsink on it..
Sep 25 01:23:03 <ds2> I favor a finger that goes between the 2 connectors and out toa  bigger dissipiator
Sep 25 01:23:07 <rcn-ee[m]> zmatt, yeap, the x15 pcb was much better designed for heat dissapation.. originaly the x15 was to have a heatsink..
Sep 25 01:24:58 <ds2> I'd think even a heatsink is optional given that it only have 2 cores
Sep 25 01:25:22 <ds2> does the am5 have multiple packages with different thermal resistances?
Sep 25 01:25:29 <rcn-ee[m]> idk, every core on the bbai, is at minimal voltage levels, the am57xx is clocked/hacked at 500Mhz, we need to find a way to shtudown the sgx..
Sep 25 01:26:16 <rcn-ee[m]> memory bus is half width..
Sep 25 01:26:19 <zmatt> my 80 degrees @ 100% load is at 1.5 GHz (cpu governor performance)
Sep 25 01:27:36 <rcn-ee[m]> my x15 is 100% 1.5Ghz, at 42.6C ;)
Sep 25 01:27:57 <zmatt> without heatsink?
Sep 25 01:29:15 <rcn-ee[m]> heatsink + 120mm fan.. sandwiched between two other boards: https://www.tinkercad.com/things/igvyVihK4NT
Sep 25 01:30:31 <ds2> actually, wonder if it you can just put a blank cape on it to duct the air through the heat sink
Sep 25 01:30:45 <ds2> use convection to move air
Sep 25 01:31:15 <rcn-ee[m]> there is a fan cape coming..
Sep 25 01:31:48 <rcn-ee[m]> you can see a picture here of it: https://www.youtube.com/watch?v=v3_hNBJbqME
Sep 25 01:32:28 <rcn-ee[m]> it was designed late, so it's more of a oh crap, we need to get this out..
Sep 25 01:32:35 <ds2> not a fan cape, just passive convection
Sep 25 01:32:37 <rcn-ee[m]> i don't have the date on that cape..
Sep 25 01:33:01 <ds2> another PCB on the connectors to create a duct and orient the opening up wards
Sep 25 01:34:41 <rcn-ee[m]> i think a big block of metal like they did on the jetson nano would work..
Sep 25 01:35:01 <ds2> I'd agree
Sep 25 01:35:26 <ds2> any benchmarks comparing TIDL on AI vs JetsonNano?
Sep 25 01:35:32 <ds2> using CUDA
Sep 25 01:36:05 <rcn-ee[m]> that's a jasonk question..
Sep 25 01:37:55 <rcn-ee[m]> (i don't have a nano or the google corral)
Sep 25 01:37:56 <ds2> I have the nano just no AI :D
Sep 25 01:37:56 <ds2> Until they fully open the corral...
Sep 25 01:38:06 <rcn-ee[m]> We could put it up against my crypto farm, 12 gtx170ti's @ 1.5KW.. ;)
Sep 25 01:38:30 <ds2> only if we adjust by the power consumption factor :D
Sep 25 01:39:01 <rcn-ee[m]> only like 10x...]
Sep 25 01:39:32 <ds2> wonder what fraction of a gtx170ti is a Jetson Nano
Sep 25 01:40:22 <rcn-ee[m]> Nano is 10Watt, 128 cores (maxwell)
Sep 25 01:41:09 <rcn-ee[m]> 1070ti is 2432cores (pascal = maxwell + 1 generation)
Sep 25 01:41:36 <ds2> and how much of that extra generation is useable for CUDA?
Sep 25 01:42:10 <rcn-ee[m]> i don't think pascal added much over maxwell for cuda
Sep 25 01:42:18 <rcn-ee[m]> just more power efficent..
Sep 25 01:42:33 <ds2> so about 1/20th
Sep 25 01:43:03 <ds2> Hmmm incremental training :D
Sep 25 02:26:45 <zmatt> jkridner[m]: so, when can we expect a beaglebone with the AM752x (J721E) ? ;-)
**** ENDING LOGGING AT Wed Sep 25 02:59:57 2019