Thread Level Parallelism

Last updated Nov 8, 2022 Edit Source

Distribute the workload among a set of concurrently running threads. Uses MIMD model.

Not to be confused with multiprocessors:

• Multiprocessors: 2 or more CPUs in the same computer. Executes multiple programs faster.
• Multicore: 2 or more processors in a single CPU. Executes a single program faster through

Overcome power wall using multiple slow cores

• Cores running at lower clock frequency and lower voltage can still deliver the desired performance using less power
• Scale up the number of cores rather than frequency

Widening gap between compute bandwidth and memory bandwidth could not be bridged - resulting in memory latency.

Overcome memory wall with memory parallelism via multiple threads