IRC Logs for #crux Wednesday, 2019-02-13

*** john_cephalopoda has joined #crux00:49
john_cephalopodaHmm01:28
john_cephalopodaSo much internet stuff is run by so few companies right now.01:28
john_cephalopodaGoogle owns youtube and basically the whole ad business. They are integrated in nearly every website via ajax.googleapis.com, maps.google.com, googletagmanager, googleanalytics or even fonts.01:30
*** xor29ah has quit IRC01:32
john_cephalopodaFacebook is embedded in various websites and also owns important social media, like Instagram and WhatsApp.01:32
john_cephalopodaTwitter is the de-facto standard for government PR. A proprietary, American service. Also they are embedded in many, many places. For many people, twitter is a news source.01:33
*** xor29ah has joined #crux01:33
john_cephalopodaAmazon owns tons of things and operates AWS, which is used by many platforms - like Netflix, Dropbox and Reddit.01:34
john_cephalopodaCloudflare is also horrible. They control a lot of the internet. And they use fucking CAPTCHAS.01:36
john_cephalopodaReplacing all those things is hard. And the hardest part is getting people to use the alternatives. When an app is more difficult to use than whatsapp, people are unlikely to change because "It's so complex, how do I do X? And why doesn't Y work? I'll just use whatsapp."01:39
john_cephalopodaYoutube has all the music. When I want to listen to something specific, youtube is basically the only source where I can find it without needing to be subscribed to anything.01:39
john_cephalopodaAnd websites just rely on cloudflare and AWS, without questioning if that's the best course of action.01:41
john_cephalopodaIf you wouldn't make your website several MB big, DDOS might be way less of an issue, for example.01:41
Anselmoyeah x-x01:45
frinnstwow, netflow from my shitty firewall totally chokes it01:51
frinnstfrom Download:    946.730 Mbit/s to Download:    554.480 Mbit/s01:53
ryuofrinnst: where can i get bandwidth that good? :(01:53
frinnstsweden :-)01:53
*** tilman_ has joined #crux01:53
*** tilman has quit IRC01:57
*** timcowchip has quit IRC02:04
*** emmett1 has joined #crux02:07
*** emmett1 has quit IRC02:08
*** emmett1 has joined #crux02:09
cruxbot[opt.git/3.4]: [notify] firefox-bin: updated to 65.0.1. Fix for MFSA2019-0402:12
*** _________mavric6 has quit IRC03:27
*** _________mavric6 has joined #crux03:28
*** emmett1 has quit IRC03:28
*** hsn has joined #crux06:34
*** dlcusa has joined #crux06:36
*** Abdullah_ has joined #crux07:10
*** Workster has quit IRC07:28
*** Workster has joined #crux07:29
*** admincomplex has joined #crux07:32
*** admincomplex has quit IRC07:33
*** admincom1lex has quit IRC07:35
*** admincomplex has joined #crux07:35
*** BurnZeZ has quit IRC08:04
*** BurnZeZ has joined #crux08:04
*** BurnZeZ has quit IRC08:06
*** BurnZeZ has joined #crux08:07
*** BurnZeZ has quit IRC08:08
*** BurnZeZ has joined #crux08:08
*** jue has joined #crux08:59
cruxbot[opt.git/3.4]: xterm: update to 34409:01
cruxbot[opt.git/3.4]: msmtp: update to 1.8.309:01
cruxbot[opt.git/3.4]: fdm: update to 2.009:01
cruxbot[opt.git/3.4]: dialog: update to 1.3-2019021109:01
*** _________mavric6 has quit IRC09:55
*** plow has quit IRC09:55
*** _________mavric6 has joined #crux09:56
*** plow has joined #crux09:57
*** guido_rokepo has joined #crux09:59
*** xor29ah has quit IRC11:22
*** xor29ah has joined #crux11:25
pedjajaeger, you are into ML stuff, this might be interesting to you https://github.com/NVlabs/stylegan12:05
john_cephalopodaMLP - Markup Language Ponies12:07
john_cephalopodaOr Machine Learning Ponies in this case.12:08
*** nwe has joined #crux12:15
*** guido_rokepo has quit IRC13:51
jaegerI've seen that one, very cool14:43
*** guido_rokepo has joined #crux14:59
*** guido_rokepo has quit IRC15:05
*** guido_rokepo has joined #crux15:24
pedjajaeger, if I am understanding ML correctly, you need beefy GPU to train, but can use pre-trained networks with something less powerful?15:55
*** uplime has joined #crux16:40
jaegeryeah, you can export a trained network's weights/biases and use them elsewhere without retraining16:50
jaegercan be very handy16:50
jaegerthere are probably many ways, that's how I'd do it with keras/tensorflow16:51
*** admincomplex has quit IRC16:52
jaegerto give you a bit of perspective, though, the bottom-of-the-barrel nvidia mx150 in my laptop smokes my threadripper 2950x in neural network training tasks16:54
pedjaI suspected there is a difference, but that much? wow17:01
pedjamakes sense, cuda render is often a couple of times faster then cpu in blender.17:02
pedjaeven thou both gpu and cpu are ancient pretty much :)17:03
*** guido_rokepo has quit IRC17:05
pedjasince I already have anaconda, setting it all up so I can play shouldn't be hard.17:06
pedjathanks, jaeger :)17:06
jaegeryeah, should be easy :)17:10
jaegernp17:10
jaegerconda install tensorflow-gpu keras <etc.>17:10
*** vlnx has joined #crux17:24
*** onodera has joined #crux18:51
jaegerpedja: http://ix.io/1AZZ <-- some timing info for a CNN on various hardware. I can't remember all the details now such as how many epochs, etc., but the important part is the relative difference between them19:27
pedjadamn, that's quite a difference19:28
pedjatesla k80? that's aws, right?19:30
pedjathat 1080ti, thou :)19:33
pedjastill a pretty bad-ass gpu19:34
jaegerYeah, the K80 was an AWS instance19:52
jaegerAnd the 1080ti is pretty awesome for these :)19:52
jaegerThe ones that are grouped together are the same system, just CPU vs GPU19:53
jaeger1: threadripper 2950x19:53
jaeger2: laptop with i5-8250u19:54
jaegerActually going to be presenting at a neural networks workshop thing tonight... same one I did last semester about iris classification20:33
pedjajaeger, that's cool :) what's the presentation about?20:42
jaegermaking a simple neural network that can classify irises in this dataset: https://archive.ics.uci.edu/ml/datasets/iris20:43
pedjapretty popular dataset, I've seen it used in a couple of data science tutorials I've watched20:45
jaegeryeah, it's a great one to start with, well-established20:45
pedja'hello world' of data science/ml :)20:46
jaeger:D20:47
*** frinnst has quit IRC20:50
*** frinnst has joined #crux20:51
*** frinnst has quit IRC20:51
*** frinnst has joined #crux20:51
pedjait's an interesting subject, but way too math-heavy to wrap my poor brain around20:55
pedjaditto for photogrametry20:56
pedjabut that one is pretty easy to use not bothering with the dark math magic behind it :)20:57
pedjasimple concept, 'recreate what the human brain and eyes are doing'20:58
jaegerSomeone else can figure out the math :)21:02
pedjayeah, bunch of phds at google hq do all the hard work :)21:03
*** xor29ah has quit IRC21:17
*** xor29ah has joined #crux21:23
*** Kruppt has joined #crux21:59
ryuoSeems the sheevaplug has a successor from GST: https://globalscaletechnologies.com/p-86-sheeva64.aspx22:11
ryuomore RAM and 64 bit now.22:11
ryuoThough, still capped at usb 2.0? the espressobin has 3.0..22:11
*** onodera has quit IRC22:52
joacimhttps://ameridroid.com/products/odroid-hc123:00
joacimwant to try this out23:00
joacimsheeva64 looks like a more complete product tho23:01
ryuojoacim: it is. i'm looking to wait for this to release.23:10
ryuojoacim: the espressobin is another one of their products.23:10
ryuojoacim: interesting, though it might be interesting to try reviving an older Marvell Orion based product.23:12
ryuoDebian still supports those for now.23:12
ryuothe HP mediavaults were based on that, but they're over 10 years old by now.23:13
*** hsn has quit IRC23:49

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!