Due to the potential to lose compute nodes in the UI queue if a job exhausts all of the node's memory, all jobs running in the UI queue will now have memory limits in place. This translates to 2G per processor core for nodes in the UI queue. If your job hits this limit the answer is usually to request more slots to spread the job out over more nodes. For a single node job, a larger memory node may need to be requested. More information about submitting large memory jobs can be found in the Advanced Job Submission Wiki entry.