Wednesday, 2023-06-14

*** TheDot <TheDot!> has joined #libre-soc00:02
sadoon[m]1<programmerjake> "latest and greatest side-channel..." <- > <> latest and greatest side-channel: power leds and rolling shutter01:21
sadoon[m]1Oh boy..01:21
*** ghostmansd[pc] <ghostmansd[pc]!> has joined #libre-soc02:11
*** openpowerbot <openpowerbot!> has quit IRC02:11
*** WhyNotHugo <WhyNotHugo!bc7d0f0b52@2604:bf00:561:2000::28> has quit IRC02:14
*** alethkit <alethkit!23bd17ddc6@sourcehut/user/alethkit> has quit IRC02:16
*** yambo <yambo!> has quit IRC02:17
*** rsc <rsc!~robert@fedora/rsc> has quit IRC02:17
*** ghostmansd <ghostmansd!> has quit IRC02:20
*** ghostmansd[pc] is now known as 042AAA2J902:20
*** ghostmansd[pc] <ghostmansd[pc]!> has joined #libre-soc02:20
*** ghostmansd[hex] <ghostmansd[hex]!> has joined #libre-soc02:20
*** 042AAA2J9 <042AAA2J9!> has quit IRC02:20
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC02:20
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc02:22
*** ghostmansd[pc] <ghostmansd[pc]!> has quit IRC02:23
*** rsc <rsc!> has joined #libre-soc02:23
*** rsc <rsc!> has quit IRC02:23
*** rsc <rsc!~robert@fedora/rsc> has joined #libre-soc02:23
*** yambo <yambo!> has joined #libre-soc02:27
*** openpowerbot <openpowerbot!> has joined #libre-soc02:31
*** alethkit <alethkit!23bd17ddc6@sourcehut/user/alethkit> has joined #libre-soc02:33
*** WhyNotHugo <WhyNotHugo!bc7d0f0b52@2604:bf00:561:2000::28> has joined #libre-soc02:36
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has quit IRC02:51
*** lxo <lxo!~lxo@gateway/tor-sasl/lxo> has joined #libre-soc02:53
*** openpowerbot <openpowerbot!> has quit IRC03:00
*** openpowerbot <openpowerbot!> has joined #libre-soc03:01
programmerjakeluke: I found out why creating a power decoder is really slow: there's a stray gc.collect() call in there! commenting that out made it go several times faster06:00
programmerjakeghostmansd: SVP64Instruction.Prefix is really slow to construct, so I had to add a cached instance. it'd be nice if it were faster and didn't need to duplicate the whole field hierarchy each time you construct an instance.07:14
programmerjakeI used py-spy with speedscope to profile running tests, which worked out nicely, once I installed libunwind-dev07:17
programmerjakeci went from 52min to 35min thanks to the optimizations i did08:07
*** ghostmansd[hex] <ghostmansd[hex]!> has quit IRC08:37
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC08:40
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc08:40
*** ghostmansd <ghostmansd!> has joined #libre-soc10:11
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC10:48
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc10:48
markos_lkcl, just got the loongson board11:28
markos_missing a case and PSU though, so will have to wait to get it powered on11:29
*** ghostmansd <ghostmansd!> has quit IRC11:41
lkclmarkos_, awesome!14:34
lkclprogrammerjake, fantastic!14:34
lkcllet me see how long it takes here, i'll time all tests in decoder/isa14:35
lkclhmm my laptop's in "powersave" mode so all cores are limited to 1.5 ghz if using parallel pytest14:37
* lkcl facepalm14:39
markos_some people misread what "eating your own dogfood" really means14:39
lkclwell it's supposed to be "fit for human consumption"14:40
lkcla fact that turned out to be really important when an Indian Restaurant in London was raided in the 1980s...14:40
programmerjakewell, training against an AI *does* work in GANs, where one neural net is producing increasingly realistic fakes and the other neural net is getting increasingly good at distinguishing the fakes from the real thing:
markos_But there is a clear separation of the roles in this case, one is trained with real data only -to generate the fakes- and the other is trained on both -to distinguish14:49
markos_the problem is when you don't know if it's a fake14:49
markos_which is the case argued in the slashdot post14:49
markos_what if you have eg. midjourney trained on its own data at one point, the results will be insane14:50
markos_they're already insanely looking images14:50
markos_but then they will make absolutely no sense at all14:50
*** psydroid <psydroid!~psydroid@user/psydroid> has quit IRC14:51
*** pangelo[m] <pangelo[m]!~pangeloma@2001:470:69fc:105::3ec5> has quit IRC14:51
*** cesar <cesar!~cesar@2001:470:69fc:105::76c> has quit IRC14:51
*** vaino[m] <vaino[m]!~vainomatr@2001:470:69fc:105::3:5474> has quit IRC14:51
*** sadoon[m]1 <sadoon[m]1!~sadoonalb@2001:470:69fc:105::3:5f7c> has quit IRC14:51
*** sadoon[m] <sadoon[m]!~sadoonsou@2001:470:69fc:105::2:bab8> has quit IRC14:51
*** programmerjake <programmerjake!~programme@2001:470:69fc:105::172f> has quit IRC14:51
*** Ryuno-KiAndrJaen <Ryuno-KiAndrJaen!~ryuno-kim@2001:470:69fc:105::14ed> has quit IRC14:51
lkclthe difference between fake-detection and training-on-training is that the fake-detection is *not* just on training-on-training, it is training on the *difference* between real and fake14:57
*** programmerjake <programmerjake!~programme@2001:470:69fc:105::172f> has joined #libre-soc14:58
*** psydroid <psydroid!~psydroid@user/psydroid> has joined #libre-soc15:19
*** sadoon[m] <sadoon[m]!~sadoonsou@2001:470:69fc:105::2:bab8> has joined #libre-soc15:19
*** pangelo[m] <pangelo[m]!~pangeloma@2001:470:69fc:105::3ec5> has joined #libre-soc15:19
*** Ryuno-KiAndrJaen <Ryuno-KiAndrJaen!~ryuno-kim@2001:470:69fc:105::14ed> has joined #libre-soc15:19
*** sadoon[m]1 <sadoon[m]1!~sadoonalb@2001:470:69fc:105::3:5f7c> has joined #libre-soc15:19
*** cesar <cesar!~cesar@2001:470:69fc:105::76c> has joined #libre-soc15:19
*** vaino[m] <vaino[m]!~vainomatr@2001:470:69fc:105::3:5474> has joined #libre-soc15:19
lkclreal    22m29.469s15:46
lkclthat's at 1.5 ghz with 8 processes15:46
lkclso... 7 minutes @ 4.8 ghz...15:46
lkcli remember it being more like 12-15 so that's damn good programmerjake15:47
programmerjakei expect it doesn't scale with frequency 1:115:47
programmerjakethough otoh it will likely go faster if it uses 16 threads on your 8 core laptop15:49
*** ghostmansd <ghostmansd!> has joined #libre-soc15:52
lkcli can't risk it overheating, otherwise i would15:54
lkclcurrently 25C and that's hot enough15:54
programmerjakeintel cpus are generally designed to be able to run at like 90C normally, and hotter when thermal throttling15:56
programmerjakethough laptops with poor cooling may be limited by max case temperature instead15:58
markos_25C? that's not hot :)16:05
markos_that's the definition of cool :)16:05
markos_my 11th gen i5 laptop is currently at 50C and I find that acceptable given the load16:07
markos_unless this measurement is reporting something else and not the cpu16:12
markos_check lm-sensors package16:12
markos_but tbh, this does not look a normal temperature of an intel cpu under load, unless your laptop is really special16:13
*** yambo <yambo!> has quit IRC16:55
*** ghostmansd <ghostmansd!> has quit IRC17:07
ghostmansd[m]programmerjake, nice catch! Thank you!17:39
*** ghostmansd <ghostmansd!> has joined #libre-soc17:49
ghostmansd[m]I think this should be hidden somewhere in Field class. I'll take a look at it later.17:53
*** ghostmansd <ghostmansd!> has quit IRC17:58
*** ghostmansd[m] <ghostmansd[m]!> has quit IRC18:07
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has joined #libre-soc18:12
*** ghostmansd[m] <ghostmansd[m]!~ghostmans@> has quit IRC18:16
*** ghostmansd[m] <ghostmansd[m]!> has joined #libre-soc18:17
*** ghostmansd <ghostmansd!> has joined #libre-soc18:31
*** tplaten <tplaten!~tplaten@> has joined #libre-soc19:28
lkclmarkos_, ahh21:06
lkclPackage id 0:  +57.0°C  (high = +100.0°C, crit = +100.0°C)  ALARM (CRIT)21:06
lkclCore 0:        +57.0°C  (high = +100.0°C, crit = +100.0°C)  ALARM (CRIT)21:06
markos_yeah, 25C looked suspiciously low21:10
*** tplaten <tplaten!~tplaten@> has quit IRC21:24

Generated by 2.17.1 by Marius Gedminas - find it at!