Lesson 135
GPU Programming & Parallel Models
SIMT · threads, blocks, grids · memory
1:00How GPUs use thousands of simple cores and the SIMT execution model to run data-parallel workloads at massive throughput.
By the end, you can
- Explain why a GPU uses thousands of simple cores rather than a few powerful ones.
- Describe the four levels of the GPU thread hierarchy and what each level shares.
- Compute a thread's global index from `blockIdx`, `blockDim`, and `threadIdx`.
- Explain SIMT and why warp divergence serialises execution.
- Identify the three levels of the GPU memory hierarchy and rank them by speed and scope.
- Describe what coalesced global-memory access is and why it matters for performance.
- Decide whether a given workload benefits from GPU execution or would be faster on a CPU.
Up next in Software Engineering & Web
