YUNOVA · Computer Organisation & Architecture
Module 12: Parallel and multicore systems
Flynn classification
Key points
- Flynn classifies computers by instruction streams and data streams.
- The four classes are SISD, SIMD, MISD and MIMD.
- Multicore processors are MIMD; vector units and GPUs are SIMD-style.
Multicore processors and cache coherence (MESI)
Key points
- Cache coherence keeps copies of the same address consistent across cores.
- MESI marks each cache line Modified, Exclusive, Shared or Invalid.
- A write to a shared line invalidates the copies in other caches.
Shared and distributed memory multiprocessors
Key points
- In shared memory, all processors use one address space.
- In distributed memory, each node has its own memory and nodes exchange messages.
- UMA gives equal access time to all memory; NUMA makes local memory faster.
Amdahl law
Key points
- Amdahl's law: speedup = 1 / ((1 - f) + f/s), where f is the enhanced fraction.
- The part that cannot be improved limits the maximum speedup.
- With 10% serial work, speedup can never exceed 10, however many processors you add.