Turn your laptop into a Deep Learning BEAST

Setup tutorial of an External Video adapter for Deep Learning. Make sure to use an Nvidia graphics card, since AMD lacks the CUDA API that most Deep Learning frameworks use.
===Steps===
1. Get the components:
+ [Mini PCI-E Version] EXP GDC Laptop External Independent Video Card Dock
+ [Nvidia] Graphic Card
+ [at least 400W] Power Supply
+ Monitor
+ WiFi dongle
2. Slip your graphics card into the PCIe slot on the BPlus board.
3. Hook your (not yet powered-on) PSU’s 24-pin ATX power supply pins into the BPlus board.
4. Connect the 8-pin PCIe connector on the board to the 6-pin power connector on the graphics card.
5. Insert the PCIe cable into the laptop, then slide the opposite side of the cable (the one with the HDMI connection) into the HDMI port labled “X1” on the PCIe adapter. At this point you'd also connect your graphics card directly to your external monitor, typically via HDMI or DVI.
6. Fire up that beast yo.
7. Disable the built-in graphics and download the drivers for your card.
8. Profit.

[@connermalloy]

You're the only one with a video on egpu's for a sensible purpose. Looking forward to seeing it do some work. Great upload.

[@dshahrokhian]

Thank you Conner.

[@manashejmadi]

wow youre using AMD cards? Good luck on that! CUDA is essential for Training models

[@Player_2725]

Basically turning a laptop into a full computer

[@igorjablan]

Very nice and useful video Daniyal Thanks for the upload! ;)

[@saltybagel7789]

say my name is jovaii georgo

[@sourabhzalke4155]

Was it a good choice ? How it is performing ? Does it worked good, are you able to perform heavy deep learning tasks in fast manner ? Please Reply !!!

[@sg6sisbro852]

Look I searched so much, it is better to sell old laptop and buy new one. Nowdays 1050 or 1050 ti can be better choice.

[@alexshi8583]

why not just use your desktop as a mainframe and use your laptop as a terminal where ever you go?

[@dshahrokhian]

That's a great alternative if you have a good desktop.

[@mikefischbein3230]

What's a good way to do that? Just use something like teamviewer?

[@wiiiiktor]

What was the overall cost? Can you paste in links of purchased hardware?

[@miguelmyers9546]

Why not just train it on the cloud and save the model

[@nautiyogi8386]

Might become costly in the long run ?

[@janewilson8029]

RECORDED ON A POTATO LESS GOOO did u eat ur camera after u filmed this video XD

[@zohaibmalik1761]

WTF MAN WHY YOUR VIDEO SHAKES LIKE THAT. THAT BLURRY VIDEO? PROBABLY GAVE ME A HEADACHE

[@SoniaReginaBordin]

can I use machine learning in game development?

[@subasurf]

Please turn youtube video stabilization off. It's unwatchable.

[@nautiyogi8386]

Question is why did you have to use the external monitor ? Didn't quite get it.

[@AskGeek]

Superb .. any way for macbook?

[@michaelxie1978]

I think the bandwidth is too low, PCIe 3.0 4x? that's only 4GB/s IO rate. This will be an issue with iterations over batches.

[@seenPunktSee]

yoo what demension u from?

[@pedrogalvez6399]

Hi! I followed your instructions but error 43 arises and my GPU (NVidia 1060) doesnt not work

[@628sonu]

how did the setup cost total?

[@furioustodd5105]

Is external monitor really required for this setup. Can I use the laptop itself as the screen?

[@dshahrokhian]

Yes you need a monitor, since the video interface in laptops is output only.

[@furioustodd5105]

Thanks. Really nice video and very helpful.

[@gauravchauhan1645]

Why not to train model in aws cloud.. It will be way cheaper..

[@srjbora]

Gaurav Chauhan How much does it cost?

[@jean4j_]

From my understanding it won't be

[@tranduy96]

way cheaper? ahahahah son have you ever tried to train anything in cloud?

[@SankhaKarfa]

Its 3.06$/hr + 0.10$/gb volume in us_east-2 for p3.2x Nvdia V100 16gb, 61gb ram

[@nautiyogi8386]

Better to use something like paperspace or floydhub instead. AWS , GC are pretty costly.Also they don't seem to have the latest GPUs.

[@omitsura]

Just buy normal desktop. video cards are more expensive parts of the build, they are just left flapping in the breeze. Slightly wrong move of wrist sends the card flying off your desk.

[@dshahrokhian]

Well, if you usually work from the same place, then a desktop is a better solution. I went with the eGPU option because I travel every now and then, and carrying an ATX with me is much more of a hassle than just a graphic card, which can easily fit in my luggage.

[@FuZZbaLLbee]

Daniyal Shahrokhian but what about the desktop power supply? Would it be possible to modify a laptop adapter from an old laptop?

[@nishantkumar3997]

Hi Daniyal if i wANT TO BUY A LAPTOP FOR DEEP LEARNING budget buy in india what would it be??

[@dshahrokhian]

I personally would never buy a laptop with an integrated high-end graphic card, since they weight like a brick and they lose the whole point of a laptop: easy mobility. I know nothing about the market in India tho.

[@riteshpatel1460]

Nishant Kumar go with desktop bro

[@8bit_hero850]

Nishant Kumar Alienware

[@surajaswin8840]

My 3GB Nvidia 970m is not sufficient. I can hardly do batch training (at max 2 images at once). That too the main problem is, my gpu doesn't run any well established architectures(More the number of parameters, more the problem using my gpu). I am fed up with resource exhausted problems .. So a better option would be a desktop with 8 or 11 GB gpu memory!

[@OEFarredondo]

Step 6 lol

[@appleolma2042]

do i need to connected to the net foe machine learning

[@ExtremeCrafter]

hi

[@truliapro7112]

Can i use GTX 1050ti ?

[@vacation9441]

will create beast like this thanks