Deep learning is an emerging technique for effective learning of discriminative features of multi-view data. Autoencoder is one of the deep models for finding single-view unsupervised tasks. Recently, multi-view clustering (MVC) techniques have been developed to assess the complementary information provided by the various data views. Thus, multi-view learning is more robust than single-view clustering or learning. There are ambiguous clustering structures in some of the views, posing poor results in MVC. State-of-the-art deep learning techniques use deep autoencoders to develop deep MVC that leverages the multi-views complementary information and constructs the global features by aggregating embedded learned features for every view. The aggregated embedded features are helpful for multi-view discriminative learning in order to create pseudo-cluster labels. Recent technique of deep multi-view collaborative clustering (DMVC), provides intra-view collaborative learning that explores the discriminative latent features for assessing the best clustering structures. Data sparsity problem occurred for the high dimensional deep features data. By the transforming the high dimensional data to lower dimensional manifolds, the effect of sparsity reduced. This paper experiments the MVS deep model for large datasets MNIST and compares the with the extended low-dimensional manifold learning based DMVC (LDM-DMVC)) results. The comparative results demonstrates that faster and accurate results are achieved with DMVC.