Blockchain-enabled Internet of Vehicles provides a reliable collaboration environment for traffic entities and promotes road safety and traffic efficiency through real-time sharings between vehicles and infrastructure. However, this is usually confronted by a bottleneck as the Internet of Vehicles scales up. In this paper, we propose a novel Graphical Consensus-based Sharding (GCS) framework, which is underpinned by four important strategies: 1) A graphical consensus is adopted as its intra-shard consensus, where the consensus group is set up according to its maximal connected subgraph, and the leader is elected by its reliability weight. The consensus group is refreshed intermittently by alternating the role of leader. 2) Within GCS, the intra-shard data are stored in the local chain, while a block-based directed acyclic graph, rather than a chained structure, is employed as the main chain. The local chain is used to respond to requests within each shard, and the main chain supports the cross-shard sharings. GCS will parallelly optimize the throughput of the blockchain-enabled Internet of Vehicles. 3) GCS further introduces the shard backup and node scheduling to handle shard failure and overheating by using new backup strategies and temporal-spatial graph convolutional network prediction model, respectively. 4) An off-chain transmission algorithm is presented for secure sharings between the infrastructure and the vehicles. Simulation results show that the number of Transactions Per Second is 1.69 times higher than that of the non-sharding blockchain, and the pending time is dramatically reduced compared to the mainstreaming sharding approach, which is 1.02s.