The 1500s errors I think always refer back to the regular sort manager and reports error in not enough tempdb buffers or proc cache buffers for the sort, the latter just used during the processing for metadata as Jeff says. Else I might have speculated it was getting throttled by the max resource granularity, but the latter only effects lava sorts ( amongst other things). Is it working out an estimate on the buffer requirement based on size of row and ultimately the row is smaller?
↧