Accelerating the LOBPCG method on GPUs using a blocked sparse matrix vector productтезисы доклада