• No results found

Current issues involved in CPU design

Output by text message

3.2.8 Current issues involved in CPU design

The control unit and arithmetic and logic unit are together called the processor. These two components are generally held on one silicon micro-integrated circuit (microchip).

The immediate-access memory is held as several microchips, which are accessible to the processor. Each memory chip may hold up to 512 megabytes of memory, depend-ing on its type. Technological developments of the CPU concentrate on two major areas:

1. speeding up the operation of the CPU, so that programs run more quickly;

2. making larger amounts of cheaper RAM available to the processor, so that a larger number of more complex programs may be held entirely in main memory during

execution. This saves the lengthy task of loading parts of programs in and out of secondary storage and main memory during a program run.

In order to achieve these objectives, several development strategies are adopted by microchip manufacturers and designers.

1. The speed of the CPU is partly determined by the clock cycle time. There is a clock in the CPU, and operations can only occur in time with the beat of this clock. If the clock is speeded up then more operations can occur per second. This is one strategy – to design microchips capable of operating with faster clock speeds. Clock speeds are measured in terms of the number of operations per second and are commonly expressed in terms of megahertz. One megahertz (MHz) is equivalent to one mil-lion operations per second. A business desktop computer might be expected to oper-ate at over 1000 MHz.

2. Increasing clock speed is not the only possibility. All processors are designed to be able to decode and execute a determinate number of types of instruction. This is known as the instruction set. It is well known that a majority of these instructions are rarely used. Microchips can be designed to operate more quickly if the set of instructions is reduced to those that are most basic and commonly used. If one of the more rarely used instructions is needed it can be carried out by combining some of the more basic instructions. These microchips are known as RISC (reduced instruc-tion set computer) chips.

3. The ALU and control unit are designed to carry out each operation with a chunk of data of a standard size. This is known as the word length. The earliest micro-chips worked with a word length of 8 bits, or one byte. If this is increased, then clearly the computer will be able to process more data in each operation. Word sizes of 32 bits are now common in business desktop computers and 64-bit processors are beginning to create interest. Mainframe computers have traditionally used longer word lengths compared to PCs. The trend is to develop processors that handle longer words.

4. Data is transferred between main memory and the processor frequently. The pro-cessor will be slowed in its functioning if it has to wait for long periods for the trans-fer of the data needed for an operation. Data is transtrans-ferred in parallel along data lines. For an early 8-bit processor (word length equals 8 bits) there were eight data lines in and eight data lines out of the processor. Clearly, one byte of data could be transferred in one operation. This coincides with the word length. Modern pro-cessors can transfer several bytes of data in each operation.

5. The speeds of operation of processors and memory chips are so fast now that, relatively speaking, a major delay occurs because of the time the electrical signal takes to move from the memory to the processor. One way to shorten this period is to decrease the distance of the processor from the memory. This is achieved by building some memory on to the processor chip. This memory is called cache memory.

6. The single processor is based on the von Neumann model of the computer. In an application, although some of the tasks must be carried out in sequence, many can be performed in parallel with one another, provided that the results of these tasks are delivered at the right times. There is considerable research and development inter-est in parallel processing using several processors, each working in parallel with the others. This can increase the rate at which the computer can carry out tasks.

7. All the developments mentioned so far are directed at increasing the speed of cessing. There is also a need to increase the amount of RAM available to the pro-cessor. This cannot be achieved simply by plugging in more RAM microchips. Each RAM location needs to be separately addressable. Memory address lines fulfil this function. If there are twenty memory address lines associated with a processor, then a maximum of 220 bytes or 1 Mb of memory can be addressed directly. Modern desktop computers may have over 512 Mb of RAM and processors are able to address increasingly large amounts of memory.

8. At the leading edge of developments, new technologies based on laser-light switch-ing and biochemically based processors are beswitch-ing investigated. These are currently in an early stage of research.

Over the last two decades, chip manufacturers have produced families of processor chips incorporating successive improvements. For instance, Intel, a major international microchip manufacturer, has produced a series of chips that are widely used in busi-ness desktop computers. The set consists of the 80X86 and Pentium processors. The later chips are improvements over the earlier ones, having faster clock speeds, larger word sizes, more data lines and more addressable RAM.

Mini case 3.7

Processors

Do not take this personally, but the microprocessor that powers your personal com-puter is pretty dumb. While the sliver of silicon can crunch gazillions of numbers per second, it can do only one thing at a time.

Until now, computer scientists have worked round this limitation using a technique known as time slicing. The processor devotes itself for a few milliseconds to, say, surfing the Internet, then moves on to burning a compact disc, then does a few of the calcula-tions that allow you to play Grand Theft Auto III.

This happens with such speed and slickness that to the user it feels seamless. The computer appears to be multi-tasking even though it is not.

But time slicing is no longer enough to satisfy the demands of users. The technology to enable microprocessors to multi-task has been around since the mid-1990s.

In contrast to time slicing, with each job dealt with in strict rotation, a ‘simultane-ous multi-threading’ (SMT) processor can work at any time on whatever requires atten-tion. It does not need to wait for the next time slice in order to change jobs.

What is more, the really big performance gains come only when the applications that the computer is working on are easily divisible into sub-tasks or ‘threads’.

This last point means that some jobs lend themselves more naturally to multi-threading than others. Serving up web pages to consumers, where each request for a page can be treated as a new thread, is the kind of application at which SMT pro-cessors excel. The order in which threads are executed is not usually critical.

In contrast, complex scientific calculations that require numbers to be crunched in strict sequence are less susceptible to the SMT treatment.

When will this technology reach consumers?

It already has. Intel’s newest Xeon chips for servers and Pentium IV chips for per-sonal computers, launched late last year, are each capable of running two

simultane-ous threads. 

IBM’s newest PowerPC processors have similar capabilities and can switch between single-threaded and multi-threaded modes as circumstances dictate.

It is Sun Microsystems, however, that is placing the biggest bet on SMT. The com-pany says that by 2006 it aims to be building computers based on chips that incorpo-rate no fewer than eight processors on a single chip, with each processor capable of running four threads simultaneously.

Adapted from: TECHNOLOGY: From soloist to symphony orchestra By Simon London

Financial Times: 26 November 2003

Questions

1. How does the SMT processor improve system performance?

2. Which applications are particularly suited to the multi-threading processor approach?

3.3 Software

‘Software’ is the general term for instructions that control the operation of the com-puter. This section deals with the basic sorts of software and the various languages in which it may be written.