TensorFlow and Keras GPU Support - CUDA GPU Setup

[@deeplizard]

👉 Check out the blog post and other resources for this video: 🔗 deeplizard.com/learn/video/IubEtS2JAiY 👀 Come say hey to us on OUR VLOG: 🔗 kzbin.info

[@john157157]

Thank you! Several days of aggravation (linux, Win7, back to linux, try again in Win7 yada, yada...) and now it's working thanks to you.

[@MrDaFuxae]

Dear deeplizard team, I love you so much for doing this videos! It took me three nights to install tensorflow with gpu support according to depency errors, missing libraries and all that stuff. I can't believe how you can manage to stay that calm while talking about this installation process for it caused me so much headache. Thanks for sharing your experience and keep going!

[@naveedbhuiyan9855]

This tutorial deserves more appreciation. i followed each step and everthing worked flawlessly. Thank you so much for this. note- if you are downloading cudnn ver 8.1.0 then there will be more than one file in the lib/bin folders. Copy and paste all of them into the other folder

[@JamesJon1187]

between all the wrong versions installed of cuda and VS, this took hours but i got it in the end. THanks for the vid!! For anyone who installed the latest visual studio and got an error message. I think the latest one nvidia cuda recognizes is 2019 or something. look it up online, but don't just go with the latest.

[@priyanshukumawat525]

your expression literally make me smile at 3:59. your lectures are truly instructive

[@valravn9773]

Thanks! This saved me lots of time fumbling around in the dark to get GPU working. Hell, I was still importing from keras until after this video when you guys informed that keras is integrated into tensorflow and that I am importing from an outdated API. Big thanks.

[@driyagon]

This channel has been my goto for a long time now. Great work guys!

[@bryanpoulter4482]

Fantastic instruction video. Very clear and the timing was great. I was able to follow along and go through the steps with you. Thanks bunches.

[@-long-]

Mild humor, charming speech covers all we need to know Hello from Vietnam (you can spot a string "Nguyen" in my name), enjoy your trip! Thanks so much.

[@sarthaksharma9322]

Most talented and beautiful lizard I ever saw! Thankyou for this amazing content!!!

[@raghavkumar6877]

Your courses are tremendous, so much clearer, crisper and to the point than the tons of courses out there. Love the short length and the industry snippets. Would love to meet you guys one day

[@ameerhussain5405]

I was about to install cuda for TF on windows and then ran into this. Thank you for making life easier👍👍 Awaiting more tutorials on this series.

[@tekingunasar8866]

Was working on a project involving style gan and was pulling my hair out over how to use GPU, very easy to follow tutorial, thanks!

[@AaronWacker]

This video is immensely useful. Great tips all around. If windows too check display adapters and version of windows to get that update: C:\ProgramData\Microsoft\Windows\Start Menu\Programs\Administrative Tools then check System Information for win version, and grab drivers. Well done and amazing video - thankyou!!

[@ThroughYourWindow]

Just starting out with all this so this was very timely. Great vid! Looking forward to delving deeper into all your content

[@khaledhanee141]

thank you for making life easier

[@Arjun147gtk]

Thanks a lot ffor this video, I have a GPU but was unaware of how to use it for deep learning.

[@Sikuq]

Thanks for this detailed CUDA setup checklist.

[@helmihelmi8904]

Nice ! I wait for a video about GAN for image data augmentation :)

[@woogonchung]

Yes. I also want to see the videos coming out. The videos for NLP and RNN things.

[@danielrosas2240]

AWESOME!!! It worked for me :)

[@harishjulapalli448]

The procedure is very long surprisingly!! I didn't had to do this for Pytorch.I could get GPU running for Pytorch within minutes.

[@Jevuify]

Thank you. Very helpful

[@adambaran-tomik6391]

Thank you, this video helped me

[@nikilson]

Wow it worked 👍 It will save my cpu❤️

[@tymothylim6550]

Thank you very much for this video! It's great to have a really beginner course for this! It helps noobies at this like me!

[@jamespaladin607]

I have found it far easier to install tensorflow using the conda terminal in Anaconda. Conda can install tensorflow up to version2.1,1, When you install using conda it automatically installs cuda toolkit version 10.1.243 and cudnn version 7.6.5.This way you do not have to do all the downloading and changing of environment variables etc. If you want to install tensorflow version 2.2 first install 2.1 with cuda the install tensorflow version 2.2 with pip as in pip install tensorflow ==2.2.0. Toolkit version 10.1.243 and cuddd 7.65 are compatible with tensorflow 2.2.

[@deeplizard]

Thanks for the tip!

[@miguelalzate4850]

Thank you. Very helpful.

[@aryan_kode]

please complete the video series on reinforcement learning. optimizing reinforcement learning

[@swanandkulkarni126]

This was extremely helpful, thank you.

[@pwnkmrdst]

As always, super cool video from super cool couple

[@rutvik3333]

Thankyou so much cheers!

[@siddharthsingh7052]

Great tutorial! thank you so much

[@turner-tune]

Amazing and helpful video, thank you very much!

[@ChrisTian-ox5nr]

This is Love! thanks for sharing!

[@pallabidas3064]

You are always the best

[@98perova]

Thanks this's been really helpful!

[@jeffnc]

Thank you, this got it working for me!

[@Matiu936]

Really helpful, finally got it to to work! Thank you!

[@herantd]

Probably the hardest part of the ai/neural network subject

[@deeplizard]

😅

[@piyushchauhan343]

thank you only your process worked

[@myrontheoharakis9392]

thank you for this!

[@azereldukali941]

Thanks a lot

[@gabrielh5105]

Thanks for this video

[@andrijanamarjanovic2212]

Bravo! I m so happy. Thank you :)

[@4sety]

Is the Game-Ready Driver incompatible or inefficient with tensorflow? Or is your choosing the Studio Driver a matter of preference?

[@rlobo8329]

great video!

[@urielvaknin6904]

First, thank you for your great job!!! Second, after following carefully all the steps in the clip I get that the number of GPUs is 0. Do you have any idea what might be the problem and how can I fix it?

[@prageethbhanuka7882]

Same issue here?

[@xSiinPZz]

Maybe you forgot to install tensorflow-gpu I was using a virtual env and was saying that my number of GPUs was 0. Issue was fixed after i installed tensorflow-gpu module and re running the code sample again.

[@dannyisrael]

You did, CUDA toolkit correct version? Cudnn correct version? Nvidia drivers correct version? Tensor flow correct version? Restart computer? There’s a table for tensorflow and all the other component versions that work together.

[@richarda1630]

Again thanks for the awesome info. Now in 2021, are there plans to make it run on Apple's new M1 ARM chipset as well? or does it do that already?

[@vinayvardhanyt2415]

front camera view is better than side view

[@nitishsonkar8947]

Thanks

[@Bill-gc9bt]

I really wish I understood all the details and lingo that you're using. I just need to figure out how to prevent my tensorflow convolutional neural network model from running out of memory. My model has about 2 million trainable parameters, and it also has an early stopping callback with a patience of 3. Everything runs fine when I call "fit" on my model. It ususally runs about 12 epochs before the early stopping callback gets activated. As soon as the callback kicks in, I get the out of memory error. I have been researching this issue for about three months, and I'm no closer to a solution path than when I started. If you could point me in the right direction towards a solution, I will be forever indebted to you.

[@mohamamdazhar6813]

Better than netflix

[@hesamshafienya7057]

I did all of the steps and still Num GPUs are 0. would you help me with this?

[@alonsorobots]

How do you enable multi GPU for tensorflow? I have two RTX2070 SUPERs and it recognizes them when I do the print line you mentioned, but when I run training it only seems to be using one core. Thanks for the fantastic video, I'm going to check out the rest of the videos on this channel now =)

[@PradHolla]

Hey what about TensorRT? Should I download it and add it's location to the PATH? Very helpful video BTW!

[@sharjeel_mazhar]

So this means that i can only use tensorflow CPU? Because I'm on a laptop and it has i guess something like Intel 620 graphic cards. Please tell me about this I'm new to deep learning and very confused by the tensorflow CPU and GPU and all about the Nvidia and all.....

[@pewpew9088]

ty

[@user-or7ji5hv8y]

What does that mean for Mac users?

[@SAsquirtle]

bruh

[@bernsbuenaobra3665]

This is a very nice easy intro years back, I did these on a Quad i7 MSI laptop. It went well, but my motherboard is reaching 100 deg.C for mesh grid simulation of a surface wave. I have since 2017 sold that laptop and acquired an ASUS ROG G751J laptop that runs on NVIDIA 970M wrote a functional CUDA cores test using CUDA FORTRAN, which also worked well. Some four years back, I ran Tensorflow code from Google, and playing with Python also went well. Now there is the latest NVIDIA Tensor AI Graphics Card. I want to buy that new GPU and use my old ROG G751J to run it like an eGPU or an external GPU. Hopefully, there is some edge connector PCIe Gen.4 somewhere!

[@bernsbuenaobra3665]

I hope you talk about NVIDIA Tensorflow AI A100 GPU's.

[@JordanMetroidManiac]

I'm trying to use Visual Studio 2019 instead of Jupyter. After hours of searching and trying different things, I _just_ can't figure out how to add the cudnn.lib file to Visual Studio (this is step 5 on the CUDNN installation guide you showed in the video).

[@eleftheriaggp4572]

i love you! Thank you really much for spreading this positive vibes combined with crazy usefull information !! I was thinking about this whole gpu installation for months now and finally did it with your help !! I was just wondering if it is possible to gain access to the gpu from other than the base environment. Currently on my device it is only finding the gpu in base-env. Somehow it would be really nice if it could be fixed also for my special "dlproject" environment. Is it possible? Did I do something wrong maybe? Thank you very much in advance. I would love to hear from you!! Greetings from Germany

[@aloysiusyeo5430]

may i know how to revert this change? it completely messed up my ide

[@shubhgaur]

You could use Conda (Anaconda/Miniconda) for setting up the environment. You won't have to go through the process of installing the CUDA Toolkit or Visual Studio or cuDNN, nor downloading any DLLs. It is as simple as -- $ conda install tensorflow tensorflow-gpu -- on a fresh installation of miniconda/anaconda. Conda handles all of the dependencies.

[@deeplizard]

Thanks for sharing your solution, Shubh. I'm not aware if there are any differences with conda's tensorflow-gpu package. I know that it does not come with TensorFlow 2.2.0 at the moment. Only 2.1.0. TensorFlow does not list the conda install on their website as an installation method.

[@shubhgaur]

Oh, okay. But I'm sure the package managers at conda will update to 2.2.0 soon. Anyway, just wanted to mention the awesome dependency solving ability of Conda.

[@shubhgaur]

@@existenceisillusion6528 Anaconda installs CUDAToolkit version 10.1.x currently. It might soon be updated to 10.2(Latest as of now).

[@4096fb]

So is it possible to use this "shortcut"?

[@debusinha9015]

Hey can you say why its showing 2 variables cuda path and cuda path v12 seperately in environment variable? thanks

[@shavkat95]

thanks a lot!

[@pierrealmhanna2226]

I downloaded CUDA 11.2 while the supported version is 11.0 should I reinstall it ?

[@rathnakumarv3956]

I am having CPU and in google for my model it have its having CUDA computing capability. how to verify this in my control panel??

[@junlinguo77]

Hi I found this really helpful ! However, I have one question, the downloaded folder contains more files (it seems like in your video, each of the three folders, bin, include and lib only contain 1 file). Should I copy everything from the downloaded folder to the corresponding directory? Thank you for your help

[@deeplizard]

Yes, copy all :D

[@junlinguo77]

@@deeplizard Hi thank you so much! I cannot explain how much help I had gained from you and the reply is so quick!

[@FanaticAuthorship]

skip to 3:40 if u want to save time

[@anishaudayakumar1778]

Amazon video! I'm currently training my mobilenet model in an environment with specification python 3.6, cuda 10.2, tensorflow 2.0.0 Additionally installing CuDNN is required? Will there be difference in the performance with or without CuDNN? Thanks in advance.

[@deeplizard]

Yes, cuDNN is required if you want to make use of the GPU while using TensorFlow/Keras.

[@josephwalker107]

Please help! Before enabling my GPU, my CNN would train like normal but now that I've enabled my GPU and tensorflow recognises it, I can no longer train? I get an output stating that it's away to start epoch 1/3 but then immediately moves on to output that it has read the dll files succesfully and never actually trains. Can anyone help me please??

[@CasualGamerJay]

Hey, I followed the steps and even got my program to recognize my GPU but when I try to OCR an image, I still get the "CUDA not available - defaulting to CPU. Note: This module is much faster with a GPU" warning. Any ideas? I'm using Tensorflow & Tensorflow GPU 2.9.0, CuDNN 8.1, and CUDA 11.2; which are all supposedly compatible.

[@AnimilesYT]

Seriously? SERIOUSLY? I have been stuck on the DLL failing to load for almost 2 whole days. I have finally fixed it by trying random things and uninstalling and installing python a few times, and it randomly started to work. Apparently my recent update to Visual Studio updated the C++ thing and after re-installing python and tensorflow again it got fixed. At least, now that I hear what you said I think that was what fixed it.. If only I'd seen this video sooner..

[@balicien]

same here

[@vacation9441]

Thanks a lot miss

[@jyotiprajapati66]

Hey I am installing tensorflow 2.4.1 can u plz tell me which version of cuda and cudnn supported

[@jerzysomkowski827]

You can also try installing it from conda from terminal: conda create -n tf tensorflow-gpu conda activate tf python >> import tensorflow as tf >> print("Num GPUs Available: ", len(tf.config.experimental.list_physical_devices('GPU'))) However this got me tensorflow gpu 2.1 and not 2.2 (which is the latest when I type this message) so I guess Conda is lagging behind a bit

[@PhantasmXD]

You, sir, are the mvp!

[@deeplizard]

Thanks for sharing your solution, Jerzy. I'm not aware if there are any differences with conda's tensorflow-gpu package, aside from it not coming with TensorFlow 2.2 at the moment, only 2.1. TensorFlow does not list the conda install on their website as an installation method.

[@sanskartewatia4320]

what to do next? it shows 1 gpu available in the terminal but how to open a jupyter notebook from this gpu? when i open the notebook from this specific environment, it shows no gpu. Man im losing my patience for real

[@tusharbarman1924]

Do you have a course on NLP?

[@deeplizard]

Not yet

[@yolandanatt9043]

what if the type of my driver is not there?, what should I do? Mine is GeForce 1080, but there's nothing like that in the list , should I keep downloading the 1660Ti?

[@SAsquirtle]

it will definitely be there

[@ronit8067]

guys have you made any videos about feature engineering? or can you guys recommend any video/channel/post for feature engineering ?

[@deeplizard]

Hey Ronit - We currently don't have any content on feature engineering nor have a specific recommendation in mind.

[@ronit8067]

deeplizard Possible future video topic ??😅

[@k9nxk654]

At 12:59 i have same error and restart not work... CUDA v10.1 set PATH as well.. any idea :)

[@ajbmathan]

Hi Mandy, thanks for the video. I followed all the steps. But tensorflow doesn't identity any GPUs. However it does identify a 'XLA_GPU' . Do you know if there is a difference between XLA_GPU and GPU. thanks [PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU'), PhysicalDevice(name='/physical_device:XLA_CPU:0', device_type='XLA_CPU'), PhysicalDevice(name='/physical_device:XLA_GPU:0', device_type='XLA_GPU')] Running tf.test.is_built_with_cuda() returns 'True'

[@vishalmethri7073]

same here

[@anujsharma8938]

at device manager in display adapters only intel(R) UHD Graphics 620 is coming and nothing like geforce or titan do i have to purchase geforce graphics card for GPU ??

[@Tiago_R_Ribeiro]

0:47 breathes!

[@shovorahman1914]

This doesn't work while i want to use tf-nightly-gpu. How to use GPU in tensorflow nightly build version?

[@improvementbeyond2974]

what if is 11.x and not a specific value?

[@DivineDusk69]

you're so beautiful Miss

[@md.imrulkayes4258]

I did everything as you say but still, it's not working. help 😥😥😥😥

[@johnsignore1615]

What if we don't have a GPU? Can we follow the course with just a CPU?

[@deeplizard]

Definitely! GPU is not required for the course.

[@unlockwithjsr]

If you have a Windows Laptop you can install Windows Subsystem for Linux which has Cuda and GPU support, they released it recently

[@HossamKorin]

Hi Mandy, I am trying to launch the notebook from my Linux docker container and at the same time have access to local folders where I keep my data and notebooks. Is there a way to do this? Thanks, Hossam

[@deeplizard]

Yes, use the --volume parameter when you run your container. See the documentation here: docs.docker.com/engine/reference/run/#volume-shared-filesystems

[@sreenivasbhaskara2692]

It will be easy to install through anaconda navigator right why to download and install so many things will there be any difference

[@ScriptureFirst]

Anaconda seemed WAAY easier, but it isn’t working in Jupyter even tho confirmed in my env & using gtx1080(well qualified)… not sure where to go from here.

[@-arabsoccer1553]

just a question,i have on my laptop ADM Radeon(TM) R5 Graphics,is i still can use GPU support or i must have NVIDIA only !?

[@deeplizard]

TF currently only supports NVIDIA cards.

[@rohitarora5411]

You can check rocmm open source library by amd for compute tasks

[@chintalapudirohitmanikanta4492]

still I'm getting 0 gpus detected, can someone help me

[@marufhasan9365]

Can I just use service like colab and avoid all these complications?

[@deeplizard]

You can follow the course with Colab. Note, you don't need a GPU to follow the course. A CPU will work fine.

[@jay-rathod-01]

sahi hai binod. bas ye hi chaihe tha.

[@tysk_dt]

Interesting. Tensorflow gpu requires CUDA 11.2, which itself requires VSCommunity 17.0. You can't get it anymore as Community Edition though. Welp...

[@arunkumar-te5cc]

how to use GPU in MAC?

[@theone3746]

I did all this and it was verified that TensorFlow can access my GPU, but I'm not seeing a speed difference. I recorded the time it took to train the network on the cpu and gpu, and they are the same 1 hour and 30 minutes. I thought the GPU would be much faster.

[@deeplizard]

Hey The One - Have a look at this one: deeplizard.com/learn/video/6stDhEA0wFQ There is a section that discusses this issue: "GPU Can Be Slower Than CPU"

[@manohariesrikrishnarajah821]

a milion likes

[@saiprakashbelkeri]

I did not have any GPU on my PC. If this is the case with you too, Just don't Install anything and use Google Colab and Do everything on Google's Cloud for free. All the required hardware will be procured in just a few seconds. Take out some time from this playlist and invest just 6 mins to Understand Google Colab. Just watch the first 2 videos from that playlist and you will be good to continue with Mandy here. Link Below. kzbin.info/aero/PLQY2H8rRoyvyK5aEDAI3wUUqC_F0oEroL Hope that helps.