Questions

1. Suppose you want to launch 100,000 threads in parallel. What is the best choice of the number of blocks in a grid and number of threads in a block and why?
2. Write a CUDA program to find out the cube of every element in an array when the number of elements in the array is 100,000.
3. State whether the following statement is true or false and give a reason: An assignment operator between local variables will be faster than an assignment operator between global variables.
4. What is register spilling? How it can harm the performance of your CUDA program?
5. State whether the following line of code will give the required output or not: d_out[i] = d_out[i-1].

6. State whether the following statement is true or false and give a reason: Atomic operations increase the execution time for CUDA programs.
7. Which kinds of communication patterns are ideal for using texture memory in your CUDA...