From the NVIDIA CUDA Development quick start guide:
CUDA-enabled GPUs have hundreds of cores that can collectively run thousands of computing threads. Each core has shared resources, including registers and memory. The on-chip shared memory allows parallel tasks running on these cores to share data without sending it over the system memory bus.
Real-time simulations of hundreds of entities immediately come to mind, sounds interesting for my next project in adm.
Check out the guide here, along with programming and SDK reference: