SIMDs are like cpus in normal computer. When you run OpenCL kernel each workgroup is placed in next available SIMD. So it doesn't mean that you can run 2 kernels. At the moment ATI doesn't support multiple kernels executions ( although it's rumored that RV8xx has hardware support for multiple kernels ).
Slide 5 states the capabilities of SIMD. SIMD is specialized for highly parallel math computations, so it quite different architecture than classical CPU. Also terminology used by Nvidia/ATI is a little bit missleading for someone coming from CPU world ( GPU thread isn't the same as thread on CPU - the closest thing to CPU thread is warp which sometimes is called hardware thread ).
I propose to read documentation from CUDA ( which has better description of general concepts ), then ATI docs.
Also here http://www.beyond3d.com/content/reviews/53 , http://www.beyond3d.com/content/reviews/52 are quite detailed information about ATI's gpus architecture ( cache size, simd specs and more ).
Retrieving data ...