Specialized Computing Domains

Most computer users are at least superficially familiar with key performance-related attributes of personal computers and smart digital devices such as processor speed and random-access memory (RAM) size. This chapter explores the performance requirements of computing domains that tend to be less directly visible to users, including real-time systems, digital signal processing, and graphics processing unit (GPU) processing.

We will examine the unique computing features associated with each of these domains and review some examples of modern devices implementing these concepts.

After completing this chapter, you will be able to identify application areas that require real-time computing and will understand the underlying concepts of digital signal processing, with an emphasis on its widespread use in wireless communication. You will also understand the basic architecture of modern GPUs and will be familiar with some modern implementations of components...

Technical requirements

The files for this chapter, including the answers to the exercises, are available at https://github.com/PacktPublishing/Modern-Computer-Architecture-and-Organization-Second-Edition.

Real-time computing

The last chapter provided a brief introduction to some of the requirements of real-time computing in terms of a system’s responsiveness to changes in its inputs. These requirements are specified in the form of timing deadlines that limit how long the system can take to produce an output in response to a change in its input. This section looks at these timing specifications in more detail and presents some of the features real-time computing systems implement to ensure timing requirements are met.

Real-time computing systems can be categorized as providing soft or hard guarantees of responsiveness. A soft real-time system is considered to perform acceptably if it meets its desired response time most, but not necessarily all, of the time. An example of a soft real-time application is the clock display on a cell phone. When opening the clock display, some implementations momentarily present the time that was shown the last time the clock display was used...

Digital signal processing

A digital signal processor (DSP) is optimized to perform computations on digitized representations of analog signals. Real-world signals such as audio, video, cell phone radio frequency (RF) transmissions, and radar are analog in nature, meaning the information being processed is the response of an electrical sensor to a continuously varying input voltage. Before a digital processor can begin to work with an analog signal, the signal voltage must be converted to a digital representation by an analog-to-digital converter (ADC). The following section describes the operation of ADCs and digital-to-analog converters (DACs).

ADCs and DACs

An ADC measures an analog input voltage and produces a digital output word representing the input voltage. A DAC performs the reverse operation of an ADC, converting a digital word to an analog voltage. ADCs often use a DAC internally during the conversion process.

A variety of circuit architectures are used in DAC...

GPU processing

A GPU is a digital processor optimized to perform the mathematical operations associated with generating and rendering graphical images for display on a computer screen. The primary applications for GPUs are playing video recordings and creating synthetic images of three-dimensional scenes.

The performance of a GPU is measured in terms of screen resolution (the pixel width and height of the image) and the image update rate in frames per second. Video playback and scene generation are hard real-time processes in which any deviation from smooth, regularly time-spaced image updates is likely to be perceived by users as unacceptable graphical stuttering.

As with the video cameras described earlier in this chapter, GPUs generally represent each pixel using three 8-bit color values, which indicate the intensities of red, green, and blue. Any perceptible color can be produced by combining appropriate values for each of these three colors. Within each color channel...

Examples of specialized architectures

This section examines some application-focused computing system configurations and highlights the specialized architectural requirements addressed in each. The configurations we will look at are:

• Cloud compute server: Several vendors offer access to computing platforms accessible to customers via the internet. These servers allow users to load software applications onto the cloud server and perform any type of computation they desire. In general, these services bill their customers based on the type and quantity of computing resources allocated and the length of time they are in use. The advantage for the customer is that these services cost nothing when they are not in use.

  At the higher end of performance, servers containing multiple interconnected GPU cards can be harnessed to perform large-scale, floating-point intensive computations on huge datasets. In the cloud context, it is straightforward and often cost-effective to break a...

Summary

This chapter examined several specialized domains of computing, including real-time systems, digital signal processing, and GPU processing. After completing this chapter, you should have greater familiarity with the features of modern computers related to real-time operation, the processing of analog signals, and graphics processing in application areas including gaming, voice communication, video display, and the supercomputer-like applications of GPUs. These capabilities are important extensions to the core computing tasks performed by the central processor, whether in a cloud server, a desktop computer, or a smartphone.

The next chapter will take a deeper look at modern processor architectures, specifically the von Neumann, Harvard, and modified Harvard variants. The chapter will also cover the use of paged virtual memory and the features and functions of a memory management unit.

Exercises

1. Rate monotonic scheduling (RMS) is an algorithm for assigning thread priorities in preemptive, hard, real-time applications in which threads execute periodically.

   RMS assigns the highest priority to the thread with the shortest execution period, the next-highest priority to the thread with the next-shortest execution period, and so on. An RMS system is schedulable, meaning all tasks are guaranteed to meet their deadlines (assuming no inter-thread interactions or other activities such as interrupts cause processing delays) if the following condition is met:

   

   This formula represents the maximum fraction of available processing time that can be consumed by n threads. In this formula, C_i is the maximum execution time required for thread i, and T_i is the execution period of thread i.

   Is the following system composed of three threads schedulable?

   Thread

   ...