Tensorflow Deep MNIST: Resource exhausted: OOM when allocating tensor with shape[10000,32,28,28]

Here is how I solved this problem: the error means that the GPU runs out of memory during accuracy evaluation. Hence it needs a smaller sized dataset, which can be achieved by using data in batches. So, instead of running the code on the whole test dataset it needs to be run in batches as mentioned in this post: How to read data in batches when using TensorFlow

Hence, for accuracy evaluation on test dataset, instead of this loc :

print("test accuracy %g"%accuracy.eval(feed_dict={ x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))

this can be used :

for i in xrange(10):    testSet = mnist.test.next_batch(50)    print("test accuracy %g"%accuracy.eval(feed_dict={ x: testSet[0], y_: testSet[1], keep_prob: 1.0}))

When i ran 1000 epochs for training and used 10 batches of batch_size = 50 for accuracy evaluation, I got the following results:

step 0, training accuracy 0.04step 100, training accuracy 0.88step 200, training accuracy 0.9step 300, training accuracy 0.88step 400, training accuracy 0.94step 500, training accuracy 0.96step 600, training accuracy 0.94step 700, training accuracy 0.96step 800, training accuracy 0.9step 900, training accuracy 1test accuracy 1test accuracy 0.92test accuracy 1test accuracy 1test accuracy 0.94test accuracy 0.96test accuracy 0.92test accuracy 0.96test accuracy 0.92test accuracy 0.94

Complementing Abhijay's answer, you can easily get the mean accuracy accross the test minibatches

accuracy_sum = tf.reduce_sum(tf.cast(correct_prediction, tf.float32))good = 0total = 0for i in xrange(10):    testSet = mnist.test.next_batch(50)    good += accuracy_sum.eval(feed_dict={ x: testSet[0], y_: testSet[1], keep_prob: 1.0})    total += testSet[0].shape[0]print("test accuracy %g"%(good/total))