Hi,
I just updated my DEM resolution topic :) Indeed there are multiple and mixed results there.
You are describing a situation which I faced before. No worries, here is what I learned :)
I have experience with high-res DEMs, and my conclusion was, that without a very good catchment area pre-processing (like Jim describes) and a solid understanding of processes and parameters, it is nearly impossible to calibrate a large area high res model. In fact, all my auto-calibration trials failed, compared to manual calibration. I think that the "standard" considerations for auto-calibration setup produces too much variability, due to the complex nature of high-res models. Simply put it, the algorithm needs exponentially more time to find a solution and most-probably just gets stuck around the non-optimal solution (phenomena described well in the SWAT-CUP manual). Coupled with long run-times for a single model simulation and you get an unsolvable task for your machine (or for your lifetime :D).
My suggestion is: either do a manual calibration or simplify the model (by HRU aggregation, DEM/LULC/SOIL map resampling, or splitting your model into smaller models).
All the best in your journey!
Natalja