โ Computer Organization & Architecture learning pathComputer Organization & Architecture ยท Module 9 of 10
Parallel processing and multicore
Doing more at once.
๐ Key ideas
- Flynn's classes: SISD, SIMD, MISD and MIMD.
- Instruction-level parallelism: superscalar and out-of-order execution run several instructions per cycle.
- Data-level parallelism: SIMD units (SSE, AVX, NEON) and GPUs process many data items with one instruction.
- Thread-level parallelism: multicore CPUs and hyper-threading.
- Amdahl's law: speed-up is limited by the part that must run sequentially.
๐ก Real-life example: Amdahl's law is like a group project: even with ten people, the part only one person can do sets how soon you finish.
SIMDSingle instruction, multiple data
MIMDMultiple instruction, multiple data
SuperscalarIssues several instructions per cycle
Amdahl's lawSpeed-up limited by the sequential fraction