That's probably because the TC value is too high and the thread is being put into dynamic GPU memory (Aka system ram, which if there isn't enough gets paged out to your hard drive - every try and calculate hashes using a hard drive for storage?? BAD!) - I had this exact some issue when I was trying to get my miner up and running and that was what I discovered.  The easiest way I figured out how to know where to set TC or buffer size was using GPU-z (people on Windows 8 should use HWInfo).  On the sensors tab, there are two graphs - Dedicated memory and dynamic memory.  Have that open when you launch your miner - if dynamic memory goes up at all, kill the mining software ASAP.  Then lower your TC until only dedicated memory increases when you launch the miner.  MY 4GB cards would do this with any value for TC over 24576 with a lookup gap of 2 (using -g 2) - buffer-size 1,536 MB.  


I don't know of any other scrypt-chacha gpu miner that supports this, so if you have a link to it, I would surely appreciate it!  (Actually, other than Vergoh's)