… Status: CUDA driver version is insufficient for CUDA runtime version

may be cuda and video card drivers are inconsistent

CUDA version with video card driver version match query: https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

Cuda version number query: nvcc-v

view graphics driver: cat/proc/driver/nvidia/version

driver version and cuda version no problem

may also be a conflict between the TensorFlow version and cuda version:

tensorflow version number query:

query TensorFlow corresponds to cuda version:

Windows side

https://tensorflow.google.cn/install/source_windows

Linux:

https://tensorflow.google.cn/install/source

final solution: lower the TensorFlow version

reference: http://www.cnblogs.com/liaohuiqiang/archive/2018/10/15/9791365.html

https://blog.csdn.net/qq_30163461/article/details/80314630

wish India and Pakistan peace, every day there are strawberries to eat

recall that in the previous section, we introduced the house-price based modeling under a single variable, the cost function, and the design of the gradient descent method. In this section, we extended the problem to the generalization, and how to deduce the gradient descent process under the change of a single variable into a multiple variable.

As shown in the figure, the input features are (x1,x2,x3,x4)

The output feature of

is y

The total number of samples

is m

It’s

and we’re assuming that the fitting function is

, then we want to minimize the cost function, the solution of the parameter is shown in the figure

in order to understand the cost function when there are only two variables, the cost function is shown in the figure

X1 = size (0-2000)

X2 = bedroom (0-5)

by observing the left figure, we find that when x1 and x2 are greatly different, the cost function image is of thin and tall shape. In this case, the route of gradient descent to the optimal solution is more tortuous, that is, the regression is more difficult. When the parameters are relatively close, as shown in the right figure, the gradient descent route is closer to a linear line, that is, the gradient descent is easier.

so we need to scale the feature

scale to [0-1] feature/ the total of number

method 2 scales to [-1, 1] (feature – mean)/the total of number

, of course, as shown in the figure, if the parameters are not far apart, such as between [-3, 3], there is no need to scale the feature

in the previous section, we talked about the effect of a on regression in the case of gradient descent, and then we specifically discuss the value of a

through the previous observation, we found that in order to find the parameters of the hypothesis function, we need to update the parameters step by step, and then introduce another solution method (through mathematical reasoning method: normal equation, one step in place to find the optimal parameter)

let’s look at this formula, X is a row vector, (XTX) is an n by n matrix, and solving its inverse requires order n3 complexity, so we don’t use this method when we have more features.

system environment :

OS: UBUNTU18.04
CUDA:10.1
Tensorflow 2.1
cuDNN: 7.6.5
TensorRT: 6.0.15(tf2.1 supports TensorRT6.0)
GPU: RTX2080(8G)*2

use the new version tensorflow (2.1 support CUDA version 10.1, 2.0 supports version 10.0), appeared the following error (error repetition code address: https://github.com/keras-team/keras/blob/master/examples/cifar10_cnn.py) : </ p>

2020-01-16 21:49:19.892263: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-01-16 21:49:19.897303: E tensorflow/stream_executor/cuda/cuda_dnn.cc:329] Could not create cudnn handle: CUDNN_STATUS_INTERNAL_ERROR
2020-01-16 21:49:19.897396: W tensorflow/core/common_runtime/base_collective_executor.cc:217] BaseCollectiveExecutor::StartAbort Unknown: 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.
	 [[{{node conv2d_1/convolution}}]]

Through the discussion of issues in tensorflow library, we know that the problem is on the video memory allocation of rtx2070/2080 graphics card. According to the method mentioned in issues, add the following code at the beginning of the program :

# gpus= tf.config.experimental.list_physical_devices('GPU')
gpus= tf.config.list_physical_devices('GPU') # tf2.1版本该函数不再是experimental
print(gpus) # 前面限定了只使用GPU1(索引是从0开始的,本机有2张RTX2080显卡)
tf.config.experimental.set_memory_growth(gpus[0], True) # 其实gpus本身就只有一个元素

but in my own context there is another error :

ValueError: Memory growth cannot differ between GPU devices

look at the hint should be the cause of conflict between gpus, so I try to use only one GPU:

import os
os.environ['CUDA_VISIBLE_DEVICES']='1' 

, this will solve the error