Classification | instruction flow | data flow | representative architecture | features |
---|---|---|---|---|
SISD | single instruction | single data | by Neumann |
|
SIMD | single instruction | multiple data | vector processor |
|
MISD | multiple instructions | single data | no widely used architecture |
|
MIMD | multiple instructions | multiple data | supercomputer |
|
It can be seen that SISD and SIMD architectures consider issues from a single control unit and single/multiple processor angles, and have strong centralized control. The MIMD architecture has the characteristics of distributed control, and multiple control units can execute different tasks in parallel, which is more suitable for complex applications.
In fact, modern computer systems have combined these architectures to form a hybrid architecture. For example, the pipeline of modern CPU also has the characteristics of SIMD; the parallel structure of MIMD is also used in GPU for graphics processing. Therefore, the framework provided by the Flynn taxonomy helps us understand computer architecture at a more fundamental level, which is the cornerstone of mastering this knowledge.
The Flynn taxonomy takes the combination of different instruction streams and data streams as the basis for classifying computer structures, and obtains four basic architectures. These four structures cover a wide range of computer architectures from early to modern times, and are an important tool for understanding the development of computer architectures.