問66
頻出度★☆☆
マルチコアプロセッサに関する記述のうち，最も適切なものはどれか。
平成25年度 秋期 ITパスポート試験 問66
テクノロジ系コンピュータ構成要素 › プロセッサ
This question's explanation and your answers so far.
解説
問 66
Incorrectア
A configuration that mounts multiple microprocessors (CPUs) in a single PC is a multiprocessor, which differs from multi-core, where a single CPU contains multiple cores. Moreover, having each processor execute the same processing at the same time and comparing the results is the idea of redundancy for detecting faults or errors and raising reliability (fault-tolerant designs such as a dual system). The cores of a multi-core processor do not duplicate the same processing; they divide up different pieces of processing, and their purpose is not reliability but improved processing capacity. Both the unit of installation and the purpose differ, so this choice is wrong.
Incorrectイ
First, the numbers themselves are wrong. A quad-core processor has 4 cores and a dual-core processor has 2, so the ratio of core counts is 4 ÷ 2 = 2 times, not 4 times. In addition, doubling the number of cores does not simply double the processing capacity. A program always retains sequential parts that cannot be parallelized (Amdahl's law), and there is also contention for memory, cache, and the bus, plus overhead from OS scheduling and synchronization, so the actual performance gain is normally smaller than the ratio of core counts. Even with the added condition that the structure of the arithmetic unit and the clock frequency are the same, the assertion of four times does not hold.
Incorrectウ
A mechanism that temporarily raises the clock frequency according to the processing load is a description of dynamic clock control (a so-called turbo or boost function), which runs above the rated value only when there is headroom in power consumption and temperature; it is not the definition of a multi-core processor itself. This function can also be implemented on a single-core processor, and conversely there are multi-core products that do not have it. The essence of multi-core is not raising the frequency but increasing the number of cores so that processing is shared in parallel, so the focus of this statement is off the mark and it is wrong.
- エ
正解 The correct answer is エ. A multi-core processor is a processor that contains, inside a single CPU (the processor package), multiple cores, where a core is a set of processing circuits that perform computation and control. A core is essentially the central part of a CPU itself; a processor with 2 cores is called dual-core, 4 cores quad-core, and 8 cores octa-core. Because each core can independently fetch, decode, and execute instructions, multiple pieces of work (threads or processes) can literally run in parallel at the same time, which increases the amount of work that can be done per unit of time, that is, the processing capacity (throughput). Choice エ states this precisely as “it has multiple processing circuits performing computation and other operations within a single CPU, each simultaneously executing different processing, aiming to improve processing capacity”: the location (within a single CPU), what it has (multiple processing circuits = cores), how it works (executing different processing at the same time), and the purpose (improving processing capacity) all match the definition. As background, it helps to remember that CPU performance used to be improved by raising the clock frequency, but the higher the frequency, the more sharply power consumption and heat increased, until this reached its limit; design therefore shifted toward increasing the number of cores and processing in parallel instead of forcing the frequency higher.
ここがポイント
- Definition of a multi-core processor: a processor that has multiple processing circuits (cores) performing computation and other operations within a single CPU, where each core executes different processing in parallel at the same time in order to improve processing capacity (throughput). Two cores is dual-core, four cores is quad-core. It spread as a response to clock-frequency increases hitting a ceiling because of heat and power-consumption constraints.
- Telling apart confusingly similar concepts: mounting multiple CPUs in one machine is a multiprocessor; executing the same processing redundantly and comparing the results is redundancy for higher reliability (fault tolerance); temporarily raising the clock frequency according to load is dynamic clock control (a turbo or boost function). Another frequently tested point is that even if the number of cores becomes n times larger, performance does not become n times greater (because of sequential parts, that is, Amdahl's law, plus resource contention and synchronization overhead).
用語