Это видео недоступно.
Сожалеем об этом.
Installing Latest TensorFlow version with CUDA, cudNN and GPU support - Step by step tutorial 2021
HTML-код
- Опубликовано: 14 авг 2024
- In this video I show you the freakishly difficult task of setting up and installing the latest tensorflow version with GPU support on Windows 10 :)
GO HERE FIRST:
www.tensorflow...
1. Microsoft Visual Studio
* visualstudio.m...
2. the NVIDIA CUDA Toolkit
* developer.nvid...
3. NVIDIA cuDNN
* developer.nvid...
4. Python (check compatible version from first link)
conda create --name tf_2.4 python==3.8
5. Tensorflow (with GPU support)
pip install tensorflow
❤️ Support the channel ❤️
/ @aladdinpersson
Paid Courses I recommend for learning (affiliate links, no extra cost for you):
⭐ TensorFlow Specialization bit.ly/3FmEVK1
⭐ Machine Learning Specialization bit.ly/3hjTBBt
⭐ Deep Learning Specialization bit.ly/3YcUkoI
📘 MLOps Specialization bit.ly/3wibaWy
📘 GAN Specialization bit.ly/3FmnZDl
📘 NLP Specialization bit.ly/3GXoQuP
✨ Free Resources that are great:
NLP: web.stanford.e...
CV: cs231n.stanford...
Deployment: fullstackdeepl...
FastAI: www.fast.ai/
💻 My Deep Learning Setup and Recording Setup:
www.amazon.com...
GitHub Repository:
github.com/ala...
✅ One-Time Donations:
Paypal: bit.ly/3buoRYH
▶️ You Can Connect with me on:
Twitter - / aladdinpersson
LinkedIn - / aladdin-persson-a95384153
Github - github.com/ala...
Duuuuuude, thank you so much!!!! You can't imagine the shear sense of hopelessness that installing Tensorflow for GPU had brought me. I once tried for almost 2 days straight with no results. But you explained it perfectly: Structured, concise and easy-to -understand !!! You're a life saver, thank you so much.!!!
you got saved bruh 😆
please can you send to me the CUDANN files because the website do not open with me please
This is the best tutorial available on the web. After having more than 2 days of sleepless nights only for figuring out the problem in installation finally watched this video. It is superb .Thanks a lot man.
Commenting 1 Year, 3 Months and 4 Weeks since this video was uploaded. The method used to ensure you are able to get the latest tensorflow still works and I was able to successfully install tensorflow-2.8.0 for python 3.8, cuDNN 8.1 and CUDA 11.2. Thank you for this video, it is the only clearest guide I found to get tensorflow to detect my GPUs. The only hiccup I faced was that the latest version of visual studio at the time of this writing (Visual Studio 2022 community edition) didn't seem to be recognized by CUDA tool kit so I had to install an older version of Visual Studio (2019) and that solved my problem. Once again, thanks a lot for this video Aladdin!
hey, how do you install the older version? i only find the version for mac and it's unavailable for windows
@@annaz2779 just google "msvc 2019"
This is awesome! This will definitely help a lot of people! I almost spent the entire Sunday last week installing my first tensorflow env with gpu support, and it was painful. At the edge of giving up and I realized that I used conda to get tensorflow gpu library and it turned out to be the cause. Always use pip install. I wish this video came early and it would save me hours of frustration.
This video's instructions works right out of the box. All steps which are needed are in the video description. And as extra feature this video told also how I can use PyCharm with created environment. Respect! If you have also Windows10 and Nvidia card, this video is all you need to get Cuda working in 30 minutes.
Can it works with nvdia geforce gtx1650 ti
jesus, why is this so complicated....the actual machine learning itself is much simpler....god this is so frustrating
I WAS HAVING THE SAME THOUGHT!! I NEED HELP 😭😭
Why? How can i help?
Yes, I hate this set up process. I just want to code and apply data science. Everything is a mess and way too complicated.
I love that you explained it in a way that allowed me use updated version. Specifically explaining how to select all the correct version numbers of tensorflow, python, cuDNN, and CUDA was the best.
Thank you so much Aladdin! I'm just getting back to coding after 20 years of doing everything else. This tutorial was 100%!
Spent 4 hours trying to configure this before this tutorial. Turns out I had conflicting versions of CUDA and tensorflow. Additionally, it appears that tensorflow 2.10 was the last version to support gpu acceleration on windows. Great video. Great explanations. Liked and subscribed.
Thanks, with out your comment I will go nuts, finally make it work after 3 hours
does it matter which windows version you have? I have 11
thank you for the guidance, I want to clarify for people installing older version: you need to specify the tensorflow version while installing (pip3 install --upgrade tensorflow==2.*.*) otherwise it will install the latest. also you need to access archive in order to download older vscode versions, also for some reason you might need to install some packages manualy in anaconda in order to run code.
Thanks a lot, after spending hours installing the correct versions and getting a lot of errors, this 9 min video is such a gem. Kudos. Saved a lot of time since then.
Thank you so much!!! I had installed cuda 11.5 and cudnn 8.3 both of which were not supported by latest tensorflow, downgraded to cuda 11.2 and cudnn 8.1 and everything works
LOL, I just yesterday went through hell figuring out the exact version of cuda, nvidia driver, pytorch and tensorflow version that supports my laptop.
Shout out you! I've been trying so many different ways of doing this and yours is the only one that worked!
Thanks for the awesome video! The only thing to keep in mind is that CUDA is currently (in Dec-2022) only compatible with Visual Studio 2019. Rest is great! Just don't waste time downloading visual studio 2022.
If I could sub 100k times I would. This video just saved me a ridiculous amount of time. Thank you so much for this video
From the bottom of my heart thanks for this amazing demonstration. I had so many bugs but after this tf gpu is running perfectly.
Got my new GPU and trying to set up CUDA with Python etc. Watched a lot of videos and read articles. Maybe it is just me, find it very hard to understand the instructions in Nvidia web. Wish someone there can rewrite the instructions for beginners. Your video is just fantastic! I will try your steps. Mind doing a quick check with Nvidia web and see if there are new steps to include as yours was done 2 years ago? If it is just the different program versions, then fine. Yours is much clearer than others, including those by some senior ML/DL youtubers. Many thanks.
I love you brother. Even though I already installed CUDA multiple times tensorflow nearly broke me. I wish you a long and healthy life
Best Video. Follow each and every instruction as he said and you will have a working GPU for any card out there. Thanks Buddy
Thank you so much for making this video. I've followed several other tutorials which all ended up not working. Your tutorial is the only one that worked.
Hi, I've reached the last part but when I run the program it doesn't say if cuda was opened succesfully or not. The program runs and the cudnn version is loaded correctly
Finally, ı read several blogs and watched several videos which all worthless. Perfect video thank you for help.
thank you thank you thank you! - I was kicking myself for buying a laptop with an MX330 (as I initially thought it was CUDA incompatible). Literally wasted tens of hours over multiple weeks f***ing around trying to get it working - managed to get it working in 30mins while watching this video. only thing is the GPU doesn't have enough memory and I'm probably better off using CPU for large batch-size tasks - but at least I know the setup process for when NVIDIA drop those tasty 4x series cards!
Its two years later and this still is a huge pain to install. I can't get it working again even with your guide, heavens knows if I'll ever figure out how I did it the first time. I recently had to wipe my computer and do a clean install of everything and my functional TF build is one of the things that I lost.
thank you!!! i was struggling so much and this was so helpful. you are a great teacher/instruction and a genius!!!
After 2 days of trying, I successfully start my training with GPU, thx bro! TF2.7 + cuda 11.2 and cudnn 8.1.
Thanks, it worked. And I had been looking over a lot websites, but your explanation was the most detailed.
Thank you so much for this tutorial! It was so easy and I was able to set this up. I just spent like 5 hours trying to get it to work, but I was able to do it in like 10 minutes with this video. Thanks!
Thanks a lot ! I come back to this video every time I want to change Cuda Version. Best explanation on the internet !
Thank you for this, tried other videos too, but hands down the joy after finally able to use my gpu for training
Perfect instructions. Not one error along the way. Thank you
If anyone else is like me and tried this in March 2024, be reaaaally specifically accurate with the version numbers they tell you on the TensorFlow site as far as the requirements if doing a Python install. Not just any CUDA or CuDNN or Python, but bang-on what they tell you. I had to jump through hoops down to minor releases, and even eventually copy over DLL files from one package to another between folders at one point. But my old computer CPU only doing 20+ min trainings turned into new computer GPU-supported *44 second trainings* so it was absolutely worth it.
Thanks a lot, I was struggling for a long time to find out how to use MX150 in python, now the problem is solved. Thanks thanks and thanks a lot.
Thank you so much for the video!
Here it is, 2023. For anyone facing a problem after following his steps, please downgrade your version of TensorFlow to a version below 2.10, such as 2.8. Besides that, you could Install Ms. VS with a version below 2022, such as MS VS 2019.
Finally!!!!
a complete and straight forward tutorial, thank you very much.
Thank you very much. It works (If have problems when installing, just manually install it and uncheck the one with problems(e.g. visual studio integration) )
Thanks man! Straight to the point and up-to-date! Love it :)
no one could explain better than this, Thank you so much
A huge shout to you... Been struggling for days
Thank you for uploading this amazing tutorial..!
An excellent demo that can be generalized to newer versions and is handy for setting up Pycharm as well.
When building the model specified in the example, I get an error.
tf.config.experimental.set_memory_growth(physical_devices[0], True)
IndexError: list index out of range.
I can't figure out what's wrong
I have the same error. Did you resolve it?
same here help please!!
i found the solution
replace these lines:
physical_devices = tf.config.list_physical_devices("GPU")
tf.config.experimental.set_memory_growth(physical_devices[0], True)
with this ones:
physical_devices = tf.config.list_physical_devices('GPU')
try:
tf.config.experimental.set_memory_growth(physical_devices[0], True)
except:
# Invalid device or cannot modify virtual devices once initialized.
pass
Thank you so much dude! Your video really helped me figuring out which CUDA and cuDNN version that I needed, it is very appreciated :)
you may be Aladdin but you did the job of a Jeanie to me, Thanx Bro!
haha, I'm glad if I saved you some time bro!
@@AladdinPersson Bro your videos are very informative and super cool, look forward to learn DL from you and please make more videos on flutter integration with tensorflow
You're a legend! Thanks for looking out for the boys!
Thank you clearest instructions on youtube for this topic !
thank you so much bro......all the documentation was so complicated and cluttered....u made it so simple!!! :)😊
This still works with TF 2.10 and Python 3.10.5 when installed with Visual Studio 2022, CUDA 11.2 and cuDNN 8.1.1. Thanks!
Thanks so much. This is very helpful. I had tensorflow working but I had to reformat and it was so annoying getting it working again but this video is a great resource.
Loads of thanks, bro. After watching so many videos finally I did it.
trying to install tensorflow on my digitalocean linux space. I am a newbie. Nice to see that I have to determine the version of tensorflow and match it with Cuda because I was getting cuda errors. That helps. Thanks
they have directions in digitalocean on doing it, but they are two years old and no longer work . Have to find another webhost. Meantime it always worked ok on my laptop. Who knew it would be do tough to deploy a tensor flow, flask app
If you guys got any errors, try pip installing tensorflow-gpu instead of just tensorflow
thank you
Worked perfectly with tf 2.6.0 (11.2.2 CUDA/ 8.1.1 cudnn) thanks for the tutorial!
What's your graphic card
@@kommanachandraganesh1080 3070
Hey Bro! Thanks for such a wonderful video and it installed successfully. I wasted one day with following one medium article. You are the saviour.Thanks man.
Yayyyy successfully installed GPU supported Tensorflow.. Thanks a ton Bro 🔥🔥🔥
Thank you!! Your tutorials are much appreciated
absolute king. spend 2 days trying to figure this shit out lol
Thanks a lot ! Finally a working setup after hours of trial and error
Thank you indeed for this step by step explanation. Very clear and simply laid out.
Thanks for the tutorial! Can you please specify why the visual studio installation is necessary? It looks like you don't use it afterwards.
Thank you man, I follow your instructions and it worked flawlessly
I need help. I did many tutorials on installing TensorFlow and i succsessfully installed parts of it, bot many data was still missing, so i repeated it, but with different CUDA Version, 11.3 but there is no cuDNN Version for 11.3, so i installed for 11.0 to 11.2. I can not reset the CURA Toolkittversion to 11.0 or 11.2. Now PyCharm says "ModuleNotFoundError: No module named 'tensorflow'". What can i do?? I`m already working 6h on this and it is not working.
same error ,did you find a way to solve it (if u remember) ?
Great explanation, and also gives an insight on how to install it, if there are any updates because I installed tensorflow 2.6, and the method still works. Love you! XD
Thank you so much for such great tutorials. I'm expecting attention and BERT implementation from scratch in Tensorflow.
Thank you bro, i would just like to add if anyone has a mistake like this
##Could not load dynamic library 'cusolver64_11.dll'; dlerror: cusolver64_11.dll not found##
Just do this:
Move to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.0\bin
Rename file cusolver64_11.dll To cusolver64_10.dll
NB: Observe your error carefully and rename to the corresponding name.
Instant like! Great Video Men! Greetings from Portugal
I appreciate you deeply from the bottom of my heart. I eventually could overcome my problems.
the best TensorFlow GPU tutorial on youtube🙌🙌
Thanks a lot. Wasted almost half a day to figure out how to make this work.
Best tutorial ever.... five star for explaining everything.
love you man, i been weeks looking around but your videosolved the issue
Great and detailed video! Thank you so much Aladdin!!
thank you , but I have issue when run code physical_devices = [] IndexError: list index out of range, there no any GPU device
Clicking on this video, one of the best decisions I've made in this week
goat!! thanks a lot new subscriber from Colombia
Please help..
I am getting this error: Could not load dynamic library 'nvcuda.dll'; dlerror: nvcuda.dll not found
Obs: I installed the current version of CUDA Toolkit (v11.4.0)
thanks dude, now i can continue my project because of you
Thanks a lot, this is a great guide, to the point. I have one question: Why are you installing Visual Studio if you are using anyhow Pycharm as an IDE.
Yeah, I too have the same doubt. Is it compulsory to install MsVS if we already have another IDE?
@@Soloist_1_ you can't install cuda without it, because cuda installation requires MSVS [2019].
This is the best tutorial ever, thanks man!!
thx for this video man. I had to change the name of 1 file at the end, but the rest worked like a charm. Only thing that is disappointing is that apparently my code runs faster on my cpu than me gpu.. apparently has something to do with calculation size or something (big data is better for gpu).
Dude! You are too awesome! Man! damn. Saved me from a great headache
Dude, that is just what I needed, thanks!
Thank you so much dude .This is best video finally it helped me a lot🙌🙌
Thank you so much for this video. Worked like a charm. Although my doubt is do i still need the Visual Studio that i just downloaded. Was that just for installation purpose?
Thanks a lot for this. Can you help me out with an error I'm seeing? I did all the steps but I'm seeing this error "Could not load dynamic library 'cusolver64_11.dll'; dlerror: cusolver64_11.dll not found". Not sure why it's looking for cusolver64_11.dll. In the video, it was looking for cusolver64_10.dll. Also, I'm using Python 3.7 so could that be an issue?
I changed the name to cusolver64_11.dll (from _10) and it ran further but got the "WARNING:tensorflow:Please add `keras.layers.InputLayer` instead of `keras.Input` to Sequential model. `keras.Input` is intended to be used by Functional model." and eventually failed with "Failed to get convolution algorithm. This is probably because cuDNN failed to initialize, so try looking to see if a warning log message was printed above."
Okay, found the issue. "Pip install tensorflow" will install latest version of tensorflow, which is 2.5.0 at this time and that was found to be the issue. I uninstalled 2.5.0 and installed 2.4.0 and it worked just like in the video!
@@V_baddict I had the same issue. Thank you very much for sharing the solution!!!
@@brunaguterres1817 no problem. Glad it helped!
Thanks a lot! this was as precise and easy as it gets
Thanks a lot bro!!.. been struggling with this for the past 2 days
I'm glad it was useful!
Thanks a lot for the tutorial!! It works like charm!
Thank you, such a great help! Much appreciated!
can I download Cudnn RNN V3 in Mac m1 Monterey ? I got the error below.
Invalid argument: No OpKernel was registered to support Op 'CudnnRNNV3' used by {{node cond_40/then/_0/cond/CudnnRNNV3}} with these attrs: [dropout=0, seed=0, input_mode="linear_input", T=DT_FLOAT, direction="unidirectional", rnn_mode="lstm", is_training=true, seed2=0, num_proj=0, time_major=false]
can you please tell me how to get rid of it
Thank you Aladdin Persson. It works now.
It worked! Thanks so much!
If I installed tf already, could I proceed to install CUDA & cuDNN or I should remove tf and start from scratch following the order you mentioned?
Yeah, I too have the same problem. I have already installed tf-gpu before, and then I proceeded to install MsVS, CUDA, cudNN according to the video. But my tf is till working on cpu. I have tired uninstalling and reinstalling tf, but it doesn't work. Please help.
thank you. after 5 attempt now i able to run the code via GPU
Thank you so much bro! This is very helpful!
Thank you very much, finally it's worked properly.
Thanks a lot for the tutorial. It worked perfectly.
Can't thank you enough man you are the absolute best 🔥🔥🔥
Thank you thank you thank you! it worked perfectly :)
great. almost spent 9 hours on installing cuda and then I found this one......!!!