Running AI on a $10,000 PC

[@guilldea]

God damn! I hope they let you keep that beast. Would that mean a 24x time reduction between uploads? Lol

[@chriscotton4207]

he didn't use his current 1080 gpu. If he did you're looking at roughly 2x faster

[@bulkyy]

@@chriscotton4207 a simple 1080 gpu wouldn't increase the performance of a 10000 pc by much at all

[@goldenblood9316]

@@bulkyy Er, no. He's referring to the fact his current PC wasn't using it's GPU where the Puget PC was using both of it's GPUs. Training utilizes the GPU a lot more than the CPU, and if you only have the CPU, it's not that fast.

[@bulkyy]

@@goldenblood9316 ahh I didn't hear him say that about his current pc

[@TheBinklemNetwork]

"okay time to render this video and" clicks "oh its done... :)"

[@Danidev]

dont let your dreams be memes

[@ThatGuyDownInThe]

my two favorite youtubers, on one page, christ almighty

[@이영훈-d7o]

Hello Dani

[@MrSpiky]

My dream is becoming a meme :D

[@thevoyageof67]

Yooooh Dani... Waduuup... I dig your stuff too

[@thevoyageof67]

@@corazon7653 🥛🍼

[@epicanimations68]

4:55 There is a warning message above it, at the top of the window. "Loaded runtime CuDNN library: 7.4.2 but source was compiled with: 7.6.0. [...] If using a binary install, upgrade your CuDNN library." The error message at the bottom says: "This is probably because CuDNN failed to initialize," So you need to update CuDNN to version >= 7.6.0.

[@mohameddharis6824]

Can't use CudNN unless you have an Nvidia gpu

[@LeviathanLegend]

@@mohameddharis6824 He has Nvidia GPUs in both systems

[@woolfel]

upgrading Cuda libraries on windows is considerably easier than on linux. But you really should use docker container or some kind of container. it doesn't help that Google's tensorflow docs on installation errors kinda sucks!

[@fabian.f97]

*tries installing a linux package on windows* Machine learning won't work on this PC!

[@DevOpsGuy0]

Tensorflow is a Linux tool????

[@edinhooliveira_]

@@DevOpsGuy0 there's a windows version too, but not all packages required are the same, recently I had a problem with some packages used with OpenCV that just had a linux version

[@DevOpsGuy0]

@@edinhooliveira_ have you worked on satellite images segmentation using unet architecture

[@wowepic2256]

how can a programmer think this lol

[@edinhooliveira_]

@@DevOpsGuy0 no, Just a simple face recognition tutorial I was following, than I discovered that one of the package used with OpenCV didn't work on Windows.

[@ianneilson]

As a Linux sysadmin this was... Relatable. But also painful. Jarvis! This is why tools exist! Do your builds in repeatable ways! Infrastructure and OS' can be code too! But, like you say, some devs just aren't in a position to learn _another_ thing, and receiving a PC that's already set up on ubuntu with the tools they need could be pretty handy. I just hope they manage to keep it up to date...

[@iamsocial4340]

Nice work Sir. Oh yea, Puget - You should definitely let Jabrils keep the demo PC - He'll change the world with it.

[@lucamedeiros90]

3:30 installing a linux TF on a windows pc doe

[@DylanBurke]

Not sure why this isn't littered all over the comments....

[@travissmith5994]

Yikes, not even running Bash on Windows!

[@lucamedeiros90]

That terminal screen is just hilarious

[@alakani]

He was probably copying and pasting from a tutorial after "sudo pip install tensorflow" failed on the previous line. Not to knock the guy, he's awesome. People tend to forget that data science and machine learning are completely separate skill sets from sysadmin and devops. It's like how many race car drivers aren't mechanics, or like how many sysadmins couldn't calculus their way out of a wet paper bag

[@fahadahmed480]

ok so that error was intentional, a oh Jabrils Lost our trust.

[@ErisGreyrat]

0:26 *Comes closer to talk telepathically quieter*

[@iambubbathedog]

This dude gonna get diabetes.

[@saiborg_]

He's literally always eating

[@sangeetavishwakarma5106]

Why he isn't fat???

[@fs-vt8kz]

It'd still be worth it.

[@dangelobenjamin]

He burn endless calories w his massive brain

[@xznmusic]

You just perfectly demonstrated the knowledge divide between I.T. and software engineers. 😂 Software engineers generally have no idea how the hardware works. I.T. generally has no idea how the software works. A funny world that we live in.

[@annawood4547]

"ladies and gents, bags and baguettes" perfect

[@remaininganonymous4069]

Didnt know of your existence till you were on safety third now i cant stop watching your vids, rapidly running out the rate im bingeing through them, really looking forward to seeing your next project keep up the awesome work!

[@separateprune717]

Next Jabrils video: "Today I'll be writing a program that files your tax return for you".

[@inconnn]

didn't you think it could've been bc the data science pc was on ubuntu?

[@bgino7567]

I like his ventriloquist skills. Snack and speak at the same time!? Fascinating.... in all seriousness. I love Jabrils energy.

[@debk.316]

Man, Jabrills should have just sent a message out to the squad, I have the .whl for that system config.

[@iammrbeat]

"Time is money." -Mr. Beat

[@quintinmakwe176]

Indeed

[@aconfusedmancx8323]

thank you for your wisdom Mr. Beat

[@Michael_Likes_Food]

MR BEAT!!!!!!!

[@cate01a]

tf r u

[@pnwscitech1589]

As someone living in the Puget Sound area, it's pronounced "Pew-jet". On another note, I love how you showed the difference and actually used a system like this for what it was intended. Poor Linus can only show video rendering and gaming benchmarks.

[@MrJoerT]

Just a PSA: you don't need expensive gear to learn programming or data science! A $50 raspberry PI will do nicely! I think this video does a bit of a disservice to inexperienced people wanting to take up development or data science. 15 orso years ago, I thought you'd need a big expensive gigahertz Pentium 3, and that my 120Mhz Pentium wasn't up for the job. But back then, my mentor showed me any old PC (or Commodore 64) would let me code hangman.

[@MrSpiky]

$10.000 is pretty expensive but for them to just give it to you... that's good for you dude

[@diegomata1062]

the first time that i actually watch the sponsor ad, dude you are entertaining great content i hope yo got to keep the pc :)

[@sarvasvarora]

Why not just use cloud GPUs/TPUs?

[@danielmacbride525]

Damn bro sick sponsors congratulations and keep up the quality vids and living your dreams

[@Dragoncro0wn]

I swear to god Jabril ate some of the packeging at 1:23

[@BrownGuyMadeIT]

so..... did they let you keep it?

[@xSaKage]

Finally a PC that's shipped with Linux :D

[@masternobody1896]

It is windows dude

[@xSaKage]

@@masternobody1896 3:58 doesn't look like windows to me

[@masternobody1896]

@@xSaKage how

[@masternobody1896]

@@xSaKage are you sure about that show me the clip

[@xSaKage]

@@masternobody1896 I already posted the timestamp, you can see during timelaps around 3:45

[@isman32]

I think you are just hilarious and a great inspiration to keep going on the AI path.

[@WhenItsHalfPastFive]

this is one of the funniest ads i've seen on youtube

[@Chasls]

Dude, been programming for 4 years now and getting a little burned out, But watching your videos makes me love it again! It inspires me to be more creative!

[@Nerd2Ninja]

Smol brain: Use a picture of yourself during live streams Big brain: Train an AI to draw a closed mouth over your face during livestreams

[@MrPicklepod]

I mean come, it could have been done using GPU's, I know it would have took longer but its kinda misleading to so CPU results next to GPU results.

[@saudfata6236]

For the mouth overlay, some post processing could do wonders! Try histogram mapping from your face to the generated image

[@codergeorge8436]

Um... running deep learning algorithms on a cpu is always going to be slower than a gpu.

[@YISTECH]

My thoughts as well

[@tyabrahamse2461]

Dude I love your videos!

[@vitr01]

I had the same problem a while ago during a computer science lecture. My laptop did not support AVX which is needed in order to run tensorflow. What I ended up doing was replacing the tensorflow backend with Theano, which supports the same instruction set as tensorflow, but is not dependant on AVX

[@electrobean]

/j I mean, it shouldn’t be that hard to process… if(mouth == open): closemouth()

[@abrarmohamed3351]

Congracts on the 10 000 000 views

[@ZEDCorbett]

"Brackeys take the wheel." Well played, sir.

[@DeepakKumar-uz4xy]

L

[@AhadCove]

Not sure what's better: The $10k pc or the Mouth Closing Algorithm 😂🔥

[@ThatGuy-qv1uu]

Why does this already have dislikes? It hasn't even started!

[@ZackAngelSmith]

@Huawa Tuam loool

[@Aw4yed]

Cause u don't know what's a fucking bot

[@clownkid]

because there's no real reason to spend over $2000 (I'd say $1500 maximum) on a PC. Technology advances quick enough where any super expensive PC you buy will be obselete in 10 years. It's just not cost effective to pay that much on a computer. 🤦

[@alexcrespo3252]

Because fucking white knight reddit virgins think that if they get the fIrSt DiSlIkE then their dad will come back

[@alakani]

@@clownkid I agree for most situations, including gaming. For programmers, artists, professional contractors - any time we spend waiting, we don't get paid. 10 year old PC?? As in, Nahalem CPU and nVidia GTS 150??? Only if I want to get fired and starve to death. As it is, it's pretty annoying having to start tasks before I go to sleep and see if they worked in the morning, on my 4x RTX 2060's that I bought used. I could get 10x more work done with some Titans, if I could afford them.

[@johndoughrealtimeeta3442]

Hay jabrills me and my baby brother watched and played a few of your games really good work Brody

[@delfininsjezus]

Good to see linux on that 10k pc

[@calebarulandu3068]

What if Brackeys, Dani, and Jabrils did a collab?

[@fuzzystudiogames]

I love his videos HOWEVER it might triggers me that he says the name wrong.... LOL Either way keep it up!

[@ByronTaaka]

It's Been Long!!! Welcome Back Jabrils. 🙂

[@you_just]

"A $10k PC? Well, it doesn't have RGB or peripherals, so it's gotta be dual Titans. And if it's AI, it's gotta be RTX for the tensor cores- yep, there it is"

[@Big_Con]

Ayyy we both got a cyberpower pc!

[@MayorAwesome]

LET THE MAN KEEP THE PC PUGET!

[@smccrode]

Puget is great. Let Jabrils keep it! Please!

[@owenleynes7086]

For old hardware you might as well put Linux on it... That gives it a chance

[@delfininsjezus]

I like your thinking. How can I employ u?

[@T-Freeman]

That's awesome. Wow I have a lot to learn. You wrote your own functional algo and I transferred star citizen to my ssd and cant load the game now. Your jedi status!

[@yasurikressh8325]

“the off the shelf computer” lol

[@derpyshark3305]

I love how he acts like he’s talking to the camera like a clog but just does voice over

[@sasuke2uzumaki]

oh snaps my bois back

[@Jabrils]

new vid next week Phillip! - Jabril

[@ghfghf7]

Were you able to train on both cards in sli with that nv bridge?

[@오재화-d6l]

Wow your content seriously changed to what KZbin like and what we like

[@optimizedpran1247]

You must have felt like the most powerful man in the world.

[@machinze_edits]

You are a great master coder brother

[@griffin7670]

Yo this guys voice could be an extremly high cocney accent, and we wouldnt know.

[@PH-x]

Yo you're rad subbed mybro love it

[@temmorijken5762]

I thought there was some weird delay in the audio but it's just a voice over

[@ahuggingsam]

next time you have trouble porting code between PCs have a look at something called Docker. It's a bit of a learning curve but its a free way to solve exactly these kinds of problems.

[@ZealothXyven]

I only came to witness the cablemanagement. Stayed because being a lazy bastard is super relatable.

[@thijsvdzwan]

"For the nerds..." You didn't mention what speed of ram you had and what type of Xeon cpu you had in there, bruh..

[@yesyes2091]

speed of ram is 2666 MHz and the Xeon cpu is a quad-core cpu. it is the Xeon W-2225. pretty bad for the money spent -_-

[@thijsvdzwan]

@@yesyes2091 lmaooo

[@Barnardrab]

In 5 years, you'll be able to buy an equivalent for $700.

[@Nathan_Daniel]

Hi Jabrils, on the screen of the laptop you planned to use (the one you ended up throwing in the bin), you were downloading the Linux version of the file. To install tensorflow without a file just run "pip install tensorflow" on the command prompt. To install with a file just click this link and save the file: files.pythonhosted.org/packages/c6/dc/9030097e5774fe02d1be3cb42eb54125d2c0607a6c11172f1dcad2b7fdcc/tensorflow-2.3.0-cp37-cp37m-win_amd64.whl By The Way: I selected this file seeing your python version in the video. Then navigate to the file in the cmd and run "pip install tensorflow-2.3.0-cp37-cp37m-win_amd64.whl" I hope this helped you Jabrils. By The Way: I love some of your AI and Machine Learning Videos. Also I am 10 years old and I am interested in AI and ML in Python.

[@sheikhskype]

They really gave you a super computer

[@mauriciolomeli2940]

You'll also help out.. yourself. LOL you got me

[@RatzTeeth]

PUGET! if you don’t let Jabrils keep this you’re missing out on major PR opportunity. I’m a novice in this world but this had made me want to educate myself to the point of investing in YOUR product! Boom that potentially 1 unit sold. Gotta spend it to make it ;) also thank you Jabril for your content it’s made me so curious about how AI and everything that comes along with works. I feel like I’m becoming apart of a something bigger looking forward “with caution” to the future.

[@duanebromfield8602]

I really hope you get to keep the PC, you deserve it!!!

[@deepaknarayan3713]

Yo Jabrils I just want to say you're videos have motivated to get started in AI. I hope I reach there soon!

[@MiyaMam948]

How's it going?

[@imalsoTOMATO]

3:40, did you ever find out how come that PC wasn't supported? I know some processors contain certain limitations when it comes to doing certain abilities, and as someone looking to maybe upgrade, I was wondering if I want to do something similar to this, would I need a new high end processor.

[@Kaka-zs4cp]

oh C'mon dude, you know the AI is not supported on your OS, you should just change the OS from windows to linux .... the off the shelf computer would run the AI

[@ivanmochalov3102]

It's nice man. Respect from Russia.

[@ferretcode]

On the low end PC you can use Anaconda to install tensorflow. I had to do it on my computer as well.

[@AshLordCurry]

My.god. 2× Sli RTX Titan? That's just beautiful to think about.

[@LaurentSparksMusic]

well... did you install the NIVIDIA CUDA 10.0 Package and delete the cud64nn.dll? cause yeah if you don't tensorflow prints that error, you need tensorflow 1.14.0 for py36 for tensorflow to work on windows 64. You also need the visual studio C++ build kit regardless of if you use VS, tf imports modules from the toolkit

[@guersomfalcon7544]

7:00 Does it take you more than 3 days to run the algorithm or to make the algorithm?

[@xscallcos4835]

u called xeon a xenon

[@ianleung2600]

dare ya to run ai on a $20 pc.

[@JamieTheEnby]

Hey jabril, as a washintonian, we pronounce puget, like Pew Jet.

[@alijahfaust]

Please Puget, let him keep it.

[@andreyrumming6842]

4:29 when my code refuses to work, then deletes itself after I make a tiny edit

[@darioushs]

6:40 Two flatten layers back to back?

[@jeffvoight6586]

Today's $10,000 PC is tomorrow's $200 PC (for some variants of 'tomorrow')

[@antoinemeloche]

@Jabrils Maybe it would have been easier to install what you needed on the laptop if you had installed Ubuntu on it like the Puget system had

[@KeeperOfKale222]

Gotta love the Goldeneye background music.

[@helpisonthewayalways]

Yeah. My name is bond Lame bond

[@user-vn7ce5ig1z]

You mean the James Bond theme? 🤨

[@helpisonthewayalways]

@@user-vn7ce5ig1z yes but specifically the goldeneye one

[@nessi3369]

He’s also got Diddy Kong’s from the snes

[@cooclamedia]

What keyboard is that? :o (at 4:42)

[@BSC4PE]

So did you get to keep it?

[@dorixcraft]

So ... installing a linux whl, on windows ... it doesn't work *pikachu surprised face*

[@tinthesin1]

I remember you from SEFD pranks, I didn't know you had this channel. I just subscribed. I remember we used to play GTA together on the Playstation, my name was NoNoSquares lol.

[@kariuki6644]

400,000,000 training parameters?! my laptop starts to get sick after 100,000 parameters

[@B_dev]

Why not lock the program to using only one core and a minimal amount of ram? Could be interesting

[@Burnthefirst]

Dude I recently went from my work laptop, a high level Asus ROG that was orignally more expensive from two years ago, to my personal self-built rig for around 2000 dollars and even that was a night and day experience. Image processing is really where the investment shows I feel. Like alone the laptop 1070 to full-size 2070 super upgrade was immense.

[@FraggaShmagga]

Jabrils I love you but this video is misleading! You're essentially just comparing CPU with GPU machine learning. It would be more accurate if you compared it to your 1080 ti.

[@tylerdurden3722]

Typical "off the shelf" PCs definitely don't have 1080tis in them.

[@DasAntiNaziBroetchen]

@@tylerdurden3722 That's right! They have a 2000-series GPU now.

[@SongStudios]

@@DasAntiNaziBroetchen Ehh. Yeeaahh....

[@infinianti]

What storage drives are you running? 6:24

[@DiogoSanti]

Hi Jabrils, any intuition on the mouth closing algorithm? Dataset with faces normal and smiling? Variational AutoEncoders forcing smile/not smile on people?

[@jordanleopold93]

But can it run Crysi-- I mean.... Grate cheese tho?

[@sirgoldman]

Dude, let me get that PC, good lord

[@debk.316]

The laptop couldn't install TensorFlow, because CPUs that do not support AVX, the TensorFlow must be version 1.5 or lower. If you want to install Tensorflow 1.5, the Python version must be 3.5 or lower. So the updated TensorFlow that was most probably trying to be installed didn't work. AVX is only on the newer CPU's and on the intel core series chips. So the cheap laptop was probably missing the AVX support.