A cluster of processing elements has a split mode in which the processing elements are configured to process independent processing workloads, and a lock mode in which the processing elements comprise at least one primary processing element and at least one redundant processing element, each redundant processing element configured to perform a redundant processing workload for checking an outcome of a primary processing workload performed by a corresponding primary processing element. A shared cache is provided, having a predetermined cache capacity accessible to each of the processing elements when in the split mode. In the lock mode, the predetermined cache capacity of the shared cache is fully accessible to the at least one primary processing element.