Building a Deep Learning PC
Рет қаралды 26,140
Күн бұрынI've been thinking about building my own deep learning PC for a while. Then two days in a row I got shut off from using GPUs on Google Cloud and it triggered a spiral into researching how I could do it.
Turns out, I needed a lot of help. Thank you Michael, Nvidia (for the TITAN RTX) and the many kind people all over the internet who put out such incredible guides.
The case came with dents in it so we called it "Denty".
Links:
Nvidia TITAN RTX GPU (what powers this beast) - nvda.ws/2X6DiuB
My blog post on the build with a bunch of helpful resources - www.mrdbourke.com/notes-on-bu...
My PCPartPicker list - au.pcpartpicker.com/list/8KXXjp
Timestamps:
0:00 - Intro and preamble
4:30 - Hardware build start
6:08 - Unboxing Nvidia TITAN RTX
8:30 - Motherboard unboxing
11:00 - CPU unboxing & install
13:00 - CPU fan unboxing & install
19:07 - RAM unboxing & install
22:12 - SSD hard drive unboxing & install
29:00 - Case unboxing (how Denty got its name)
35:00 - Power supply unboxing & install
46:00 - First power on
47:20 - Nvidia TITAN RTX install
52:40 - Second power on
53:51 - Software installation begins
54:55 - Installing Ubuntu
56:16 - Installing Miniconda and Nvidia drivers (CUDA & cuDNN)
56:37 - Software installation test
59:26 - Running first deep learning model on Nvidia TITAN RTX
59:55 - How I plan on using the new deep learning PC
1:01:22 - Outro timelapse
Connect elsewhere:
Get email updates on my work - dbourke.link/newsletter
Support on Patreon - bit.ly/mrdbourkepatreon
Web - dbourke.link/web
Quora - dbourke.link/quora
Medium - dbourke.link/medium
Twitter - dbourke.link/twitter
LinkedIn - dbourke.link/linkedin
@marvindesmond7697 3 жыл бұрын
Daniel's excitement while holding the Titan RTX ... Deep Learning to another level !!! Now I expect more interesting videos coming up...
@beloaded3736 3 жыл бұрын
Very infomative as always, thanks Daniel.
@piyushtalele3594 3 жыл бұрын
You seriously are a mentor to lot of guy's like me..An inspiration to move forward to one's goal to become a AIML developer. Thanks a lot for your efforts. Keep inspiring 🤘🏻🤘🏻🤘🏻🤘🏻
@mrdbourke 3 жыл бұрын
Keep getting after it Piyush!!!
@yogeshhasabe4931 3 жыл бұрын
GREAT to hear you back after such a long gap...🙌 Daniel.. Phenomenal piece of creation been displayed here..!!
@mrdbourke 3 жыл бұрын
Thank you Yogesh!
@a10m1c 3 жыл бұрын
Please make videos more often🥺 I always wait for you to come up with something informative everytime 🥺🥺
@mrdbourke 3 жыл бұрын
Plenty more videos to come
@marvindesmond7697 3 жыл бұрын
If there was a "pip install every-deep-learning libray" Daniel would have gone for it...
@innocentiuslacrim2290 3 жыл бұрын
Behind this video was a great gesture by Nvidia on the Titan card :-) I am doing an update on my own computer, but I have been thinking of going into either 3950X or even the Threadrippers to get from 16 to 64 CPU cores. I already have an RTX 2070 so I will keep that. I was a bit surprised you went for "just" 8 cores, but that is plenty of power compared to my old overclocked i5-6600K running at 4,7 GHz. You probably need more GPU cores anyhow and your card is a beast on that front. Much of the data analysis I do does not run on GPUs at all so the CPU is really important for me. Getting a new PC that you build by yourself is such a great feeling. Congrats on your build! Your friend is a great guide on your computer build if he has not that much build experience (I understood so from the early comments). It is pretty fun to listen to someone building a computer who clearly has not been working that much on the hardware side before. You are (were) pretty clueless on that side :-D but when you moved to the software side your comfort level kept on increasing exponentially and pretty soon you were on your area of expertise. From here on it is "smooth" sailing :-D Have fun!
@anirudhnallamalli6527 3 жыл бұрын
This was really helpful, I'm on my own PC building journey!
@fatemehcheginisalzmann2189 3 жыл бұрын
Thanks, I was waiting for this video :)
@mrdbourke 3 жыл бұрын
Awesome! Let me know if you've got any questions
@alexeysytnikov2529 3 жыл бұрын
Wow! Daniel, I've found you here, on KZfaq. I'm so impressed by your ideas that you deliver in your videos and share them completely. I'm so grateful for the course you've done with Andrei on Udemy. Your presentation and delivery are just perfect. Thank you for your work! Keep on going! I suppose I'm going to leave a suscribe button grey for a very long time:)
@mrdbourke 3 жыл бұрын
Hey Alexey! Thank you so much my friend, that’s a massive compliment. So stoked to hear you’re enjoying the course! Thank you for subscribing, I really appreciate it.
@rahulpawar904 3 жыл бұрын
Enjoyed watching it brother. it was really fun build. hypnotized by Fan bruh
@mrdbourke 3 жыл бұрын
Thanks Rahul! Haha that fan clip turned out better than I thought!
@dayaljaiswal6832 3 жыл бұрын
Thanks you Guru
@Qtc64 3 жыл бұрын
So good man
@Michael-London 3 жыл бұрын
Yep i did what your doing about 2 years ago for the same reasons and bought a 2080ti - no relying on cloud.,,easy to debug / step through code locally....and no server costs...in England the heat produced was also good for heating that flat! I want to upgrade to the new 30 series!
@ocixot94 3 жыл бұрын
At my job they got a TITAN RTX last year. They asked me to investigate about how to make it run. I saw a post where they use it as an external GPU using a Razer Core X enclosure. What would you recommend me. To go for the enclosure or to build a workstation?
@mrdbourke 3 жыл бұрын
Workstation can hold more GPUs I believe. But I don’t have that much experience with external enclosures. Might require a bit of research
@rutuparnpawar1022 Жыл бұрын
I love the build. I am building a PC for ML stuff with Nvidia Tesla K80.
@hepengye4239 3 жыл бұрын
Do you have any suggestions of whether a beginner should start with win10 environment or Linux env?
@mrdbourke 3 жыл бұрын
Whichever you’re most comfortable with. I’ve only used macOS for the last ~8 years so Linux feels a bit better to me. Plus, I don’t need many of the things Windows offers. But that’s my use case
@hojinryu5177 3 жыл бұрын
I just started learning deep learning on Mac. I am not sure if I can actually work with this MacBook Pro to study deep learning . any advice pls? Do I need to buy one? or actually build one also.?
@mrdbourke 3 жыл бұрын
Use your MacBook Pro until you can’t You’ll get very far with Google Colab
@psyno1856 3 жыл бұрын
"If you are gonna be coding python, you've got be rocking pythons " lmao
@mrdbourke 3 жыл бұрын
🐍💪
@su-1337 3 жыл бұрын
@@mrdbourke didn't thought will find you here
@SAGARSHARMA-cd8pj 3 жыл бұрын
How big was the dataset? 11 s/epoch for how much data computation (with resolution) ? Also, can I get the code and dataset link to measure my laptop perfomance?
@mrdbourke 3 жыл бұрын
Relatively small, 750 training images @ 224x224 with 2500 validation images @ 224x224, 32 batch size. Haven’t got the code available yet (replying on phone). Let me make a Colab notebook for you when I get home and I’ll share it
@mrdbourke 3 жыл бұрын
all code in this colab notebook is what I used in the video: colab.research.google.com/drive/1Q-q5p5yqLZd0g3Vi5s_AMUikw4uY6cZ3?usp=sharing should be able to copy it and use it locally you'll need TensorFlow installed
@pavelsgrjozs2377 3 жыл бұрын
Ps/ And how about graphic drivers on Linux for Nviidia products? As I understand this driver what exist on Linux aren't provided by Nvidia, they are provided by Linux repo supported community.
@mrdbourke 3 жыл бұрын
I haven’t tried to download any of these, I won’t be using the PC for many graphic intensive purposes, mostly numerical coding
@parthchokhra948 3 жыл бұрын
Why did you use titan RTX ? Wait for Ampepre 3090?
@valdius85 3 жыл бұрын
It's free. NVIDIA is happy to give away a relatively old product when they already announced a new one. It is great marketing for them and they helped a good YT out. Beneficial for both parties.
@mrdbourke 3 жыл бұрын
Waldemar is right. 3090 looks beast but a free Titan RTX is plenty enough for me for now. When I push this thing to its limits, I’ll look towards the 30 series line.
@valdius85 3 жыл бұрын
@@mrdbourke I have similar style. I'll be looking for second hand 1060ti when all the rich kids start to replace their toys :)
@LegendarySkyHero 3 жыл бұрын
How did he get it for free?
@codingeinstein983 3 жыл бұрын
Nice
@kirtipandya4618 3 жыл бұрын
Can you please list your configuration and the parts you have used? Thank you 🙏🏽
@mrdbourke 3 жыл бұрын
Check the description for the parts list link!
@jeffrey5602 3 жыл бұрын
great vid as always, would have recommended to start with 2 x 32 GB sticks though. Leaves more upgrade options later on.
@mrdbourke 3 жыл бұрын
Ooo, great suggestion. I didn’t even realise you could go that high!
@jeffrey5602 3 жыл бұрын
@@mrdbourke taught by experience -.-' i have a threadripper platform with 8 ram slots and unfortunately started out with 4 x 8 GB because that looked massive 2 years ago. Sold those and bought used 4x16. Now I am unsure if i should have gone for 2 x 32GB to leave room to go to 256GB sometime in the future which TR supports. especially with the bump in vram with 3090 approaching. This is gonna be a bad year for my wallet :)
@dhruvmewada4303 3 жыл бұрын
Hey I have doubt about the processer Amd or intel
@mrdbourke 3 жыл бұрын
I’m not an expert here, my advice would tend to be: choose the one you can afford
@mahdishirmohammadi3066 3 жыл бұрын
is Titan RTX better than RTX 3090 for deep learning?
@mrdbourke 3 жыл бұрын
I'm not sure on this one, this post might help out: timdettmers.com/2020/09/07/which-gpu-for-deep-learning/
@satria5403 3 жыл бұрын
What do u think with upcoming RTX 3000 series?
@mrdbourke 3 жыл бұрын
Incredible, worth looking into if you’re thinking of building a deep learning/gaming rig
@AselIndula 3 жыл бұрын
what is the best GPU for deep learning? TITAN RTX or RTX 3080?
@mrdbourke 3 жыл бұрын
both are great, RTX 3080 doesn't have as much memory as TITAN RTX though, however, it's probably just as fast at computing - but I don't have official numbers, I'd check Nvidia's website for those
@AselIndula 3 жыл бұрын
@@mrdbourke thank you!
@atharvahude 3 жыл бұрын
Which laptop should I get for Deeplearning? I cannot affort expensive hardware . I am a student.
@udaylunawat2163 3 жыл бұрын
As he said, use google colab if you're a student or just starting out. The GPU'S on colab are sufficient enough
@atharvahude 3 жыл бұрын
@@udaylunawat2163 thanks 😊
@mrdbourke 3 жыл бұрын
Uday is right, buy the computer you can afford and write lots and lots of code. If you need a GPU, use Google Colab, those GPUs are world class.
@davidedwards7172 3 жыл бұрын
Maybe my setup might help you with some setup. I use a 13" MBP for my main computer. In my closet I have 1 dell r730 with 2x K80's and an r720 with 1x K80. I use my MB to hammer out the code, and test to make sure it with launch without error. Then You can uses ssh or sfp to transfer everything over, or I use Filezilla to transfer the files over. You then can manually run the scripts via ssh, or setup a watch folder to auto run the scripts when a new one arrives into the folder(this is what I do)
@mrdbourke 3 жыл бұрын
Woah, thank you for the great tips David. Taking notes on this.
@davidedwards7172 3 жыл бұрын
@@mrdbourke You are welcome. I have a lot more experience in the hardware department of deep learning than I do with the software side(working on that currently). On thing I would really want to try with a newer GPU like yours is Mixed Precision. www.tensorflow.org/guide/mixed_precision. It seems pretty straight forward and is the best way to use the Titan RTX Tensor Cores. The K80s I have do not have the FP16 ability. They have Better FP64 that are needed for some workflows(bioinformatics, oil and gas). You should notice a 2.5x to 3x speed up in train using Mixed Precision and a reduction in VRAM usage.
@willykitheka7618 3 жыл бұрын
That's at 48:35 when you're peeling off the plastic🤣🤣🤣🤣🤣
@gero8049 3 жыл бұрын
I'm surprise how little you know about mounting a pc and boot up. Gaming really teaches you a lot in this area.
@mrdbourke 3 жыл бұрын
This was the first PC I built!
@KenJee_ds 3 жыл бұрын
🔥
@mrdbourke 3 жыл бұрын
😘
@XX-vu5jo 3 жыл бұрын
Sadly, I am using a gtx 1070 which is waaaay less than the cost of that titan. Yet, it runs almost similarly the same lol.
@martinbadoy5827 3 жыл бұрын
I bet they took the design of the Cooler Master CPU cooler mount from a motor skills IQ test :p
@neural1023 3 жыл бұрын
try monitor in vertical mode
@theperfectionist1607 3 жыл бұрын
You should try Linux Manjaro. It's better optimized for Nvidia GPUs.
@ChaiTimeDataScience 3 жыл бұрын
The true secret to Deep Learning is RGB, you need to add more RGB 😁
@davedrives212 3 жыл бұрын
That hat is freaking sickkkkkkkkk
@mrdbourke 3 жыл бұрын
Thanks bro. Some Chad got it for me
@jd_flick 3 жыл бұрын
Pugent just released benchmarks on the 3080 and apparently they are beating the titan RTX now. Crazy.
@mrdbourke 3 жыл бұрын
Holy crap. I need to checkout these out.
@arranwalker2647 3 жыл бұрын
The damage to you case is most likely due to shipping, from experience working at Amazon I have noticed you have to throw sh*t around to keep up with the packages coming down the line if you don't you fall behind. So people working in shipping throw sh*t.
@mrdbourke 3 жыл бұрын
Hahaha I know where you’re coming from - I used to work at the airport and packages would get thrown around a lot. Karma comes around!
@pavelsgrjozs2377 3 жыл бұрын
Hi - Why Intel why not AMD ?
@mrdbourke 3 жыл бұрын
No reason other than the Intel chip was a good deal
@MohammedAli-dm8le 3 жыл бұрын
Please do a setup reveal.
@Erickeagle16 3 жыл бұрын
How much it cost
@Erickeagle16 3 жыл бұрын
super duper sir
@piyushtalele3594 3 жыл бұрын
Comes to around 5.5 lkhs as per INR see the pcpartpicker page for reference
@Mirsche95 3 жыл бұрын
go to 11:07 there you can see the pcpartpicker list and the total price :)
@eric-theodore-cartman6151 3 жыл бұрын
Instructions unclear , I built a Cyborg at home......😂😁
@mrdbourke 3 жыл бұрын
Hahahaha... I hope it’s a friendly cyborg!
@ashikshafi1765 3 жыл бұрын
Wow brother you can do now live stream with Tinnie 😂👍get some COD
@mrdbourke 3 жыл бұрын
Machine learning is my favourite game!
@akolchanov83 3 жыл бұрын
3090 must have
@mrdbourke 3 жыл бұрын
I agree! They look epic!
@HarshithMK 3 жыл бұрын
Who wants to tell him about the 3090?
@mrdbourke 3 жыл бұрын
It looks like a great card!
@kaminenianirudh 3 жыл бұрын
Bro, you don’t need an RTX titan to be running these ML models. Cuz if they are too big, then chances are that you’ll have to run it on a remote cluster anyway. Don’t misguide people. Nobody buys a GPU (especially one as costly as an RTX titan) to run neural network models. You’d usually take an Amazon EC2 subscription to run it. Much more cheaper that way.
@kaminenianirudh 3 жыл бұрын
One idea for a future video is to review how to use a remote cluster for running your big ass neural network model and maybe review the different subscription tiers.
@kaminenianirudh 3 жыл бұрын
Let me elaborate before somebody finds my comments offensive. The reason why people don’t have crazy local rigs to run NN models is because the RAM is usually the limiting factor in local setups. If you are running Neural network models which are running fast enough with 32 gigs of RAM (like in the above video), then I’m pretty sure that the model is small enough not to warrant any GPU acceleration (if it does, even an 499$ RTX 2070 would be overkill, much less a 2500$ Titan). If you are dealing with huge datasets (usually the kind of problems which require GPU acceleration are ones with 100 GB datasets), you’re better off subscribing to a data service provider. Data centers give you options to choose from 1000Gigs of RAM and GPU solutions which are specific to Data Crunching like NVIDIA DGX A100 (not media/graphics related card like an RTX titan). This would be the preferred way if you wanna run a crazy enough model which requires hardware acceleration(like retraining the OpenAI’s GPT-2 or something). Otherwise boys, you can do pretty much everything that the rig here can do, on a simple RTX 2070 too (with maybe a 1.5x overhead). I’d rather buy a 500$ laptop and take an appropriate EC2 subscription (or any other) when I wanna run my model.
@davyshah1074 3 жыл бұрын
just a month after , NVIDIA dropped a bomb
@mrdbourke 3 жыл бұрын
They sure did! Hahaha
@edcunningham5562 3 жыл бұрын
Haha I love Australians
@jazz1tup 3 жыл бұрын
The content is nice, but you should get a gimbal for your videos, this is a pain to watch.
@mrdbourke 3 жыл бұрын
Thank you for the feedback
@snippletrap 3 жыл бұрын
Didn't bother me at all. Fun vid
@vikramc.n.1121 3 жыл бұрын
Hey Daniel sir, hope you are safe and good. I'm Vikram, I took your ZTM machine learning course, I wanted to learn more in-depth on machine learning so I thought I can follow your self created AI masters degree but my parents were forcing me to do a masters degree in college. so, I saw an online masters degree program and I actually registered to that program but I'm really confused about whether it will be good for my career or let me just follow your self created AI masters degree. So, can you tell me whether this is a good option, below is the link for that, since you are an experienced person, your guidance for me will be really useful and gives me a clear vision. link - www.upgrad.com/masters-in-ml-ai-ljmu-iiitb/
@dhirajpatil1602 3 жыл бұрын
I would suggest not to go for upgrad! They are just making use of ml hype to rob people. If you are really intrested do the free courses such as fastai or stanford ml course. And if you are more intrested in research side get a copy of deep learning book by Ian Goodfellow. Dont think there is more to learn than this. Afterwards you can participate in kaggle competition to get a job
@vikramc.n.1121 3 жыл бұрын
@@dhirajpatil1602 Oh, thank you so much for your advice Dhiraj sir, I registered for it because my parents were forcing me to do an MSc degree course also I paid 35000, now I am really confused whether to do it or to do an MSc degree in abroad somewhere. I explained my father about self study but he is not even listening to me. Anyway I'll try to explain him and thank you so much sir.
@mrdbourke 3 жыл бұрын
Hey Vikram. I have no experience with an actual master’s degree so I can’t comment on what its like to do one. But put it this way: all of the best resources for ML are available online in some way. That means if you can dedicate yourself in some way, create your own curriculum, work on projects and share your work (through GitHub or your own blog or both), you can get a very valuable education and hands-on experience for a fraction of the price. Use the projects you’ve worked on to share your skill with potential employers. This article may help you: www.mrdbourke.com/how-can-a-beginner-data-scientist-like-me-gain-experience/
@vikramc.n.1121 3 жыл бұрын
@@mrdbourke Thank you so much, Daniel sir for guiding me. I'll explain it to my parents and I'll try to study on my own. Also, your course was really awesome, explained well, and waiting for another course from you with unsupervised learning, clustering, association rule learning, reinforcement learning, and a lot more in it. Thank you, sir.
@1ycx 3 жыл бұрын
Uhh, buy 3090, or atleast 3080 cuz it's cheaper and better than 2080 Ti
@kirangouds 3 жыл бұрын
Instead of TITAN you could waited a few days and get RTX 3090
@mrdbourke 3 жыл бұрын
You’re right, the RTX 3090 looks awesome
@gowthamgopinath8003 3 жыл бұрын
Why did you buy the rtx titan v now......you could have waited for the rtx 3090 it's faster than the rtx titan
@valdius85 3 жыл бұрын
He got the card for FREE.
@mrdbourke 3 жыл бұрын
You’re right! The 3090 looks great but Waldemar is also right, Nvidia were kind enough to send me the Titan RTX for free
@valdius85 3 жыл бұрын
@@mrdbourke I'm happy they did. I meant it in a positive way. They offered you an outgoing product that is still valid and expensive. Without that you might had struggled to have that PC. It is good for both of parties.
@TeoZarkopafilis 3 жыл бұрын
TPUs > Titan RTX
@koby9340 3 жыл бұрын
computuh
@luckspell 3 жыл бұрын
9/1/91 is my birthday whats yours?
@mrdbourke 3 жыл бұрын
1st September 1993! Are we on the same day??
@luckspell 3 жыл бұрын
@@mrdbourke yep mines cool eh? :)
@mrdbourke 3 жыл бұрын
@@luckspell it sure is, 9, 1, 9, 1... in Australia we write dates day/month/year, so mine turns out to be 1/9/93, which is 1993, the same year I'm born haha - high five for epic birthdates!
@wolfisraging 3 жыл бұрын
Why bro why :( why did you use titan rtx when new series is just released... you could have waited till new cards are available... and at that point this titan rtx could have easily be bought for about half the price... or you could have just buy the new 3090 which is the beast of its kind and way better than titan rtx .... and that too at lesser price...
@kevinmccallister7647 3 жыл бұрын
Nvidia sent it to him for free man....
@mrdbourke 3 жыл бұрын
You’re right Wolf, the new 30 series looks amazing! But Kevin is also right, Nvidia sent me a Titan RTX for free, hard to turn down that offer. It’s a phenomenal card. Once I’ve pushed it to its limits, I’ll look towards a 30 series
@wolfisraging 3 жыл бұрын
@@mrdbourke Yeah if that's the case then yes you made good decision :)
@kevinmccallister7647 3 жыл бұрын
@@mrdbourke you can have two graphics cards right? The one you have now and a 30 series?
@ShaikAsad 3 жыл бұрын
First
@mrdbourke 3 жыл бұрын
LEGEND!!!
@prathikshenoy9158 3 жыл бұрын
@@mrdbourke lol
@ShaikAsad 3 жыл бұрын
@@prathikshenoy9158 im always first
@bhoopensawant8269 3 жыл бұрын
Bro you should have waited for RTX 3070, double the performance for half the price
@mrdbourke 3 жыл бұрын
The new 30 series looks epic! But as for this card, Nvidia were kind enough to send me this one to try out.
@bhoopensawant8269 3 жыл бұрын
@@mrdbourke Holy smokes !! Getting an RTX for free is amazing !!
Eater Straw Hat
Рет қаралды 29 МЛН
Jeff Heaton
Рет қаралды 44 М.
Jeff Heaton
Рет қаралды 63 М.
Linus Tech Tips
Рет қаралды 764 М.
Buy Smart┃Магазин ноутбуков
Рет қаралды 1,7 МЛН