Title: GPU. Lesson 2. GPU Architecture and Parallel Processing

Title: GPU. Lesson 2. GPU Architecture and Parallel Processing

1. GPUs feature a massive array of cores for parallel processing.

2. A GPU core handles a single computational thread in parallel.

3. Thread refers to a sequence of instructions executed by cores.

4. A warp, in GPU terms, is a group of threads.

5. SM, or Streaming Multiprocessor, manages warps within GPU architecture.

6. Memory hierarchy includes registers, shared memory, and global memory.

7. Registers are small storage units within each GPU core.

8. Shared memory allows fast data sharing among threads in SMs.

9. Global memory connects to all cores, storing large data sets.

10. Latency, the delay in data access, affects GPU efficiency.

---

Technical Examples:

1. Warps and Threads: A warp consists of 32 threads in NVIDIA GPUs, executing together for efficiency.

2. Streaming Multiprocessor (SM): Each SM allocates resources to manage multiple warps concurrently.

3. Memory Hierarchy: Registers hold immediate data, shared memory aids thread collaboration, and global memory stores bulk data.