which PyTorch version is correct? - python

So, I had downloaded CUDA 11.2 for tensorflow keras, which worked great, but now I want to use pytorch for reinforcement learning, and currently pytorch only has two options for CUDA version for pip downloads, 10.2 and 11.3
If possible I don't wanna upgrade to cuda 11.3 or downgrade to 10.2, without doing that is it possible to download pytorch for a cuda version that is not recommended for?
Like will 10.2 recommended pytorch work for cuda 11.2 version

https://github.com/veritas9872/PyTorch-Universal-Docker-Template
Please try out this repository. It contains a template to build any version of PyTorch on any version of CUDA, cuDNN, Python, etc.
Please star it if you find it useful!

Related

Can I install different versions of CUDA toolkit and PyTorch?

If my CUDA version is 11.2 with PyTorch CUDA version 10.2, will it cause a problem if I later install NVIDIA CUDA toolkit 11.3?
no you can not do this.
the libcudnn version need corresponding to correct cuda version
try to use cuda 11.3 for pytorch
look this answer: I can't train my NN with TensorFlow using GPU

tensorflow 2.4.1 doesnt detect gpu

I'm kind of new to machine/deep learning. I installed TensorFlow versions 2.4.1, I have CUDA version 11.2 but and cudNN when I want to get a list of available GPUs it returns nothing.(my GPU is 1050 ti 4GB)
I tried to install tensorflow-gpu but nothing changed.
what should I do?

Why my CUDA 10.1 is identified as CUDA 11.0 when trying to install Tensorflow in Windows 10?

I am trying to install Tensorflow 2.2 (or later) in Windows 10. According to the official Tensorflow instalation guide, Python 3.8 support requires TensorFlow 2.2 or later. I installed Anaconda with python 3.8 and then tried to install tensorflow using conda install -c anaconda tensorflow but it displays 2 errors:
My Python version is not compatible (although the tensorflow page says the contrary).
My CUDA version is 11.0 (but I installed the 10.1 version as specified in the tensorflow installation guide).
In this picture I show the message errors
Additionally I tried using only conda install tensorflow but it displays the same messages as before.
I also tried doing the installation both inside and outside the enviroment I created named sstensorflow but it doesn't work.
Regarding the second error message, I used nvcc --version to check the installed version of the CUDA driver and it says it is version 10.1 as shown in this picture.
So I don't know why my computer admits having CUDA 10.1 but when trying to install tensorflow it says I have CUDA 11.0 and also I don't know what is the error regarging my python version. Please help me.
I had a similar problem. Had to go back to python 3.7. Other issue is that when it says
Your installed version is 11.0 I believe it is referring to your GPU card driver not the CUDA version. I had to find a driver version compatible with CUDA 10.1. I have an RTX 2070 GPU and the driver version I have is 26.21.14.3200. GO to the Nvidia site and search for a driver for your GPU card that is compatble with CUDA 10.1

Tensorflow shows only "successfully opened CUDA library libcublas.so.10.0 locally" and nothing about cudnn

My tensorflow only prints out the line:
I tensorflow/stream_executor/dso_loader.cc:152] successfully opened CUDA library libcublas.so.10.0 locally when running.
Tensorflow logs on the net has lots of other libraries being loaded like libcudnn.
As I think my installation performance is not optimal, I am trying to find out if it is because of this. Any help will be appreciated!
my tf is 1.13.1
NVIDIA Driver Version: 418.67
CUDA Version: 10.1 (I have also 10.0 installed. can this be the problem?)
According to TensorFlow documentation, cuDNN is a requirement for tensorflow-gpu. If you don't have cuDNN installed, you wouldn't be able to install tensorflow-gpu since the dependency library would be missing.
So, if you have successfully installed tensorflow-gpu and are able to use it, e.g.
import tensorflow as tf
tf.Session()
you are fine.
EDIT
I just checker here and tensorflow_gpu-1.13.1 officially only supports CUDA 10.0. I would recommend to use it instead of CUDA 10.1.
Further, NVIDIA recommends using driver version 410.48 with CUDA 10.0. I would stick with it as well.
Actually i always rely on a stable setup. And i tried most of the tf - cuda - cudnn versions. But most stable was tf 1.9.0 , CUDA 9.0, Cudnn 7 for me. Used it for too long without a problem. You should give it a try if it suits you.

How to use Tensorflow GPU with CUDA 10 on RTX 2070?

in a few days I will setup my new computer with a RTX 2070.
I would like to user tensorflow GPU but I can't find compatible versions of CUDA and Tensorflow GPU.
As far as I know, I need CUDA 10 to benefit from the additional computing power of the RTX's Turing architecture. But regarding to the Tensorflow website the newest version of tf (tensorflow_gpu-1.12.0) only works with CUDA 9.
I would prefer to get it all working on windows 10 but if there is no other way, linux would work as well.
Somewhere on the internet I read about two rumors:
1. there is some way to compile an unpublished version of tf-gpu which works with CUDA 10
2. they will publish an official version of tf-gpu in january 2019 (which is almost over now) which will support CUDA 10.
Can someone confirm one of those rumors (with source would be the best) or tell me how I will be able to get it all working?
You're correct that you need cuda 10 and that tensorflow-gpu currently doesn't support it. What you need to do is compile tensorflow from source like your first rumor.
Installation steps:
Install CUDA 10 and cuDNN 7.3.1
Configure Tensorflow and compile it
Install the .whl package with pip
Here are some tutorials to compile tensorflow.
Windows:
https://www.pytorials.com/how-to-install-tensorflow-gpu-with-cuda-10-0-for-python-on-windows/2/
Ubuntu:
https://medium.com/#saitejadommeti/building-tensorflow-gpu-from-source-for-rtx-2080-96fed102fcca
https://towardsdatascience.com/how-to-make-tensorflow-work-on-rtx-20xx-series-73eb409bd3c0
Alternatively
you can find the pre-built tensorflow wheels here, thus skipping step 2:
https://github.com/fo40225/tensorflow-windows-wheel

Categories

Resources