The first thing to look out for would be if you have enough RAM to store all the embeddings. It's

num_dim * 7 * num_documents + normal RAM

. Apart from that, the larger the number of dimensions of the model, the more computational operations needed to perform the KNN search