Section 3.1. Stand-alone Processors

3.1. Stand-alone Processors

Stand-alone processors refer to processor chips that are dedicated solely to the processing function. As opposed to integrated processors, stand-alone processors require additional support circuitry for their basic operation. In many cases, this means a chipset or custom logic surrounding the processor to handle functions such as DRAM controller, system bus addressing configuration, and external peripheral devices such as keyboard controllers and serial ports. Stand-alone processors often offer the highest overall CPU performance.

Numerous processors exist in both 32-bit and 64-bit implementations^[2] that have seen widespread use in embedded systems. These include the IBM PowerPC 970FX, the Intel Pentium M, and the Freescale MPC74xx Host Processors, among others.

^[2] 32-bit and 64-bit refer to the native width of the processor's main facilities, such as its execution units, register file and address bus.

Here we present a sample from each of the major manufactures of stand-alone processors. These processors are well supported under Linux and have been used in many embedded Linux designs.

3.1.1. IBM 970FX

The IBM 970FX processor core is a high-performance 64-bit capable stand-alone processor. The 970FX is a superscalar architecture. This means the core is capable of fetching, issuing, and obtaining results from more than one instruction at a time. This is done through a pipelining architecture, which provides the effect of multiple streams of instruction simultaneously. The IBM 970FX contains up to 25 stages of pipelining, depending on the instruction stream and operations contained therein.

Some of the key features of the 970FX are as follows:

• A 64-bit implementation of the popular PowerPC architecture

• Deeply pipelined design, for very-high-performance computing applications

• Static and dynamic power-management features

• Multiple sleep modes, to minimize power requirements and maximize battery life

• Dynamically adjustable clock rates, supporting lower-power modes

• Optimized for high-performance, low-latency storage management

The IBM 970FX has been incorporated into a number of high-end server blades and computing platforms, including IBM's own Blade Server platform.

3.1.2. Intel Pentium M

Certainly one of the most popular architectures, x86 in both 32- and 64-bit flavors (more properly called IA32 and IA64, respectively) has been employed for embedded devices in a variety of applications. In the most common form, these platforms are based on a variety of commercial off-the-shelf (COTS) hardware implementations. Numerous manufacturers supply x86 single-board computers and complete platforms in a variety of form factors. See Section 3.2, "Integrated Processors: Systems on Chip," later in this chapter for a discussion of the more common platforms in use today.

The Intel Pentium M has been used in a wide variety of laptop computers and has found a niche in embedded products. Like the IBM 970FX processor, the Pentium M is a superscalar architecture. These characteristics make it attractive in embedded applications:

• The Pentium M is based on the popular x86 architecture, and thus is widely supported by a large ecosystem of hardware and software vendors.

• It consumes less power than other x86 processors.

• Advanced power-management features enable low-power operating modes and multiple sleep modes.

• Dynamic clock speed capability enhances battery-powered operations such as standby.

• On chip thermal monitoring enables automatic transition to lower power modes, to reduce power consumption in overtemperature conditions.

• Multiple frequency and voltage operating points (dynamically selectable) are designed to maximize battery life in portable equipment.

Many of these features are especially useful for embedded applications. It is not uncommon for embedded products to require portable or battery-powered configurations. The Pentium M has enjoyed popularity in this application space because of its power- and thermal-management features.

3.1.3. Freescale MPC7448

The Freescale MPC7448 contains what is referred to as a fourth-generation PowerPC core, commonly called G4.^[3] This high-performance 32-bit processor is commonly found in networking and telecommunications applications. Several companies manufacture blades that conform to an industry-standard platform specification, including this and other similar stand-alone Freescale processors. We examine these platforms in Section 3.3, "Hardware Platforms," later in this chapter.

^[3] Freescale literature now refers to the G4 core as the e600 core.

The MPC7448 has enjoyed popularity in a wide variety of signal-processing and networking applications because of the advanced feature set highlighted here:

• Operating clock rates in excess of 1.5GHz

• 1MB onboard L2 cache

• Advanced power-management capabilities, including multiple sleep modes

• Advanced AltiVec vector-execution unit

• Voltage scaling for reduced-power configurations

The MPC7448 contains a Freescale technology called AltiVec to enable very fast algorithmic computations and other data-crunching applications. The AltiVec unit consists of a register file containing 32 very wide (128-bit) registers. Each value within one of these AltiVec registers can be considered a vector of multiple elements. AltiVec defines a set of instructions to manipulate this vector data effectively in parallel with core CPU instruction processing. AltiVec operations include such computations as sum-across, multiply-sum, simultaneous data distribute (store), and data gather (load) instructions.

Programmers have used the AltiVec hardware to enable very fast software computations commonly found in signal-processing and network elements. Examples include fast Fourier Transform, digital signal processing such as filtering, MPEG video coding and encoding, and fast generation of encryption protocols such as DES, MD5, and SHA1.

Other chips in the Freescale lineup of stand-alone processors include the MPC7410, MPC7445, MPC7447, MPC745x, and MPC7xx family.

3.1.4. Companion Chipsets

Stand-alone processors such as those just described require support logic to connect to and enable external peripheral devices such as main system memory (DRAM), ROM or Flash memory, system busses such as PCI, and other peripherals, such as keyboard controllers, serial ports, IDE interfaces, and the like. This support logic is often accomplished by companion chipsets, which may even be purpose-designed specifically for a family of processors.

For example, the Pentium M is supported by one such chipset, called the 855GM. The 855GM chipset is the primary interface to graphics and memorythus, the suffix-GM. The 855GM has been optimized as a companion to the Pentium M. Figure 3-1 illustrates the relationship between the processor and chipsets in this type of hardware design.

Figure 3-1. Processor/chipset relationship

Note the terminology that has become common for describing these chipsets. The Intel 855GM is an example of what is commonly referred to as a northbridge chip because it is directly connected to the processor's high-speed front side bus (FSB). Another companion chip that provides I/O and PCI bus connectivity is similarly referred to as the southbridge chip because of its position in the architecture. The southbridge chip (actually, an I/O controller) in these hardware architectures is responsible for providing interfaces such as those shown in Figure 3-1, including Ethernet, USB, IDE, audio, keyboard, and mouse controllers.

On the PowerPC side, the Tundra Tsi110 Host Bridge for PowerPC is an example of a chipset that supports the stand-alone PowerPC processors. The Tsi110 supports several interface functions for many common stand-alone PowerPC processors. The Tundra chip supports the Freescale MPC74xx and the IBM PPC 750xx family of processors. The Tundra chip can be used by these processors to provide direct interfaces to the following peripherals:

• DDR DRAM, integrated memory controller

• Ethernet (the Tundra provides four gigabit Ethernet ports)

• PCI Express (supports 2 PCI Express ports)

• PCI/X (PCI 2.3, PCI-X, and Compact PCI [cPCI])

• Serial ports

• I2C

• Programmable interrupt controller

• Parallel port

Many manufacturers of chipsets exist, including VIA Technologies, Marvell, Tundra, nVidia, Intel, and others. Marvell and Tundra primarily serve the PowerPC market, whereas the others specialize in Intel architectures. Hardware designs based on one of the many stand-alone processors, such as Intel x86, IBM, or Freescale PowerPC, need to have a companion chipset to interface to system devices.

One of the advantages of Linux as an embedded OS is rapid support of new chipsets. Linux currently has support for those chipsets mentioned here, as well as many others. Consult the Linux source code and configuration utility for information on your chosen chipset.