ALPHA processor

ALPHA Architecture

Alpha Architecture February 25, 1992, a meeting in the meeting in Tokyo was officially promoted, the key features of the new architecture were all listed . At that time, Alpha was just the internal code of product development. The new processor uses a complete 64-bit RISC design to perform fixed length commands (32 bits). There are 32 64 Bit integer registers that operate 43-bit virtual addresses (later can be extended to 64-bit). As in VAX, use the Little-Endian byte order, that is, the low-byte register occupies the low memory address line. Unlike most processors such as Motorola, that is, low byte registers take up high memory address lines. In addition, the processor also built an arithmetic coprocessor, with 32 float 64-bit registers, with random access, instead of the stack access method used on the Intel X86 coprocessor. The life cycle of the entire alpha is designed for at least 25 years.

Alpha's streamlined instruction set

The simplified instruction set is more conducive to the pipeline operation, which is composed of 5 parts:

· Integer instruction

· floating point directive

· branch and comparison instructions

· Read and Storage instructions

· PAL encoding instruction

21264 chip maintains an Alpha processor to run multiple operating systems, including Tru64Unix, OpenVMS, and Linux, etc., there are many mature applications, which already have many mature applications. It is also an advantage of the Alpha processor.

For users who pursue performance, DEC's Alpha processor is a tempting choice. Alpha is the fastest in the RISC processor, and is the only RISC processor that has been supported by WindowsNT (1995, NT interrupt support for MIPS and PowerPC). If your application efficiency is limited by several high-end applications, these apps can run on alpha, then Alpha may be a good choice for you.

Today's Alpha chip is 21164, with a variety of models from 300MHz to 600MHz. In an integer application, the performance advantages of Alpha have not been very large compared to Pentiumii, although its clock frequency is very high, but the average operation per cycle is less. In fact, whether the advantage of the Alpha processor can be embodied in the application, in various applications that require floating point operations, such as 3-D image processing and computer aided design, 21164 can display superior performance.

In the technical aspect, 21164 has an amazing design of up to 9.6 million transistors. It has a 16K-level cache integrated in the chip and a 96K secondary cache that is also integrated within the chip. The width of the system bus is 128, which has a larger bandwidth compared to the 64-bit bus of the X86 system.

For all RISC processors, the biggest challenge in the PC market is to lack application software. Although Digital is trying to attract software developers and there is a very amazing application list, these applications are mainly facing engineers and creators, rather than mainstream PC applications.

To solve this problem, Digital has developed an interesting software called FX! 32, which allows Alpha to run almost all 32-bit Windows applications. Performance is better than other software simulators, such as SoftPC, but it only uses 50% to 70% of Alpha native code (Digital's statement). FX! 32 is a good way to run those complete software suits, which can meet your needs of X86PC, but this does not provide leading high performance and value. Most Alpha systems are much more expensive than typical PC systems.

Digital offers a new processor called 21164PC, which is much cheaper than 21164 and the performance is similar to. 21164PC has 400, 466 and 533MHz models. The price of this new processor is approximately $ 495 (533MHz model) at 1000 pieces, compared to the 500MHz 21164 price at $ 1450, and 233MHz Pentiumii price is $ 386.

Digital hopes that the system of using this new chip will appear at the end of 1996, and the price is only $ 2,500. With this low-cost system, Alpha can occupy a part of the PC market - if there is enough to attract the user's application. In 1998, Digital plans to issue 21264, which allows Intel's fastest X86 processor to be eclipsed, especially in floating point operations. By 1999 or later, systems using this chip may fall to about 10,000 US dollars. If the price is not a problem for you, and you want to get the fastest WindowsNT system in the world, this processor is your choice.