hewlett-packard UNITED STATES
Skip site-wide navigation link group hewlett-packard home products and services support solutions how to buy
hewlett-packard logo with invent tag line - jump to hp.com home page
End of site-wide navigation link group
printable version
digital technical journal online
hp labs skip lorem ipsum dolor navigation menu link group
contact hp
table of contents
online issues
hp journal home
hp labs home
about hp labs
news and events
careers @ labs
technical reports
worldwide sites
end of lorem ipsum dolor navigation menu link group
foreword - Volume 7 Number 1

CURRENT ISSUE - Volume 7 Number 1 Richard L. Sites,
Alpha CPU Co-architect,
Corporate Consultant Engineer

Welcome to the Tenth Anniversary issue of the Digital Technical Journal! As Jane explains in the Editor's Introduction, much has changed since Volume 1 Number 1. In fact, much has changed in only three years, as this issue on Alpha attests.

From Race Cars To Express-Delivery Trucks

What if race cars evolved as quickly as computers? They might stay on the race track and get about three times faster every three years. Or they might move off the track into everyday life.

Just three short years ago, Digital was celebrating in these pages and elsewhere the first-generation Alpha hardware and software announcements. The typical analogy likened Alpha to a race car--blazingly fast, but not seen in your own neighborhood. In this issue, Digital's engineers describe second-generation Alpha hardware and software. Today's analogy is to an express-delivery truck--still faster than the rest of the industry, but now so commonplace in the delivery of business data that it is almost taken for granted.

If we look back at history, most healthy computer architectures did not become firmly established in the marketplace until the introduction of their second generation. Following are architecture generations for three popular processor designs, with milestone implementations highlighted. Each architecture goes through many generations, on about a three-year cycle. The second generation in each case has been the point at which the architecture became firmly established. Subsequent generations have represented continuing refinement and performance increases. Alpha fits this industry-wide pattern.

                    IBM System/360    VAX             Alpha
                    --------------   -----------      ------------
Architecture design 1962             1975             1989
First generation    1965 360/40      1978 VAX-11/780  1992 21064 
Second generation   1968 360/85      1982 VAX-11/750  1995 21164 business
Third generation    1971 370/145     1985 MicroVAX    [1998 21264]
Fourth generation   1974 Virtual     1988 VAX 6000    :
                         memory           Model 200  
Fifth generation    1977             1991 VAX 6000
                                          Model 600  
Sixth generation    1980             1994
Seventh generation  1983 3090        :
Eighth generation   1986 
Ninth generation    1989 ES/9000
Tenth generation    1992 
So, with the advent of second-generation Alpha computers, how are we doing?

At The Starting Line

The first-generation Alpha hardware designs focused on performance, especially on high clock rate for the CPU chip. The initial 150-MHz and 200-MHz Alpha systems entered the market when the Intel 486 at 66 MHz and MIPS R4000 at 75 MHz were the fastest chips of their respective lines.

The first-generation Alpha software focused on survival--enough operating system functionality (VMS subset) to support at least some customers, and enough compiler optimization to make at least some (single-user, Fortran, scientific) programs run fast. An important part of the introduction was migration software to help the installed base of VAX/VMS and MIPS/ULTRIX customers move to the Alpha platform. As with any new venture, the approaches were minimalist, with sophistication left to the future.

Like race cars, the initial Alpha hardware and software was criticized for being temperamental--fast indeed for some applications, but not appropriate for others. Some observers assumed that the first generation was a fluke, or that it represented all that Alpha computers could ever be. The reaction to many architectural features, such as 64-bit addressing or relaxed read-write ordering, was "who needs it?" They wrote Alpha off as a niche design.

Delivering On Its Promise

Three years later, the future has arrived with a (muffled) roar:

  • The sophisticated second-generation Alpha 21164 chip, described in this issue, is 1.5 to 2 times as efficient as the first-generation 21064 chip in terms of work done per clock cycle on real programs.

  • The chip clock rate has been boosted from 200 MHz to a stunning 300 MHz.

  • Efficiency of compiled code has improved by 10 to 60 percent on many programs.

  • Operating system code has been expanded and tuned.

The performance factors roughly multiply together, producing second-generation systems that are about 2.5 to 3.5 times faster than the equivalent first-generation systems. One example of the higher speeds of these new systems is the AlphaServer 8400, discussed in this issue; system performance approaches the level of supercomputers with Linpack nxn results of 5 GFLOPS.

The second-generation system platforms emphasize industry leadership for a broad range of commercial client-server applications, not just scientific applications. Like express-delivery trucks, much of the second-generation software is focused on enterprise-wide database access. Truly taking advantage of the 64-bit addressing for the first time, Oracle 7 database software can run huge in-memory database queries 200(!) times faster than traditional 32-bit database software. The three database papers in this issue emphasize Digital's focus on commercial applications.

Operating system support is substantially more robust and has been expanded to the fastest UNIX and Windows NT implementations in the industry. Full VMS clustering, including mixed Alpha and VAX clusters, is available. UNIX and NT clustering is announced. All three operating systems now support SMP, symmetric multiprocessing. The 64-bit Digital UNIX implementation has led the rest of the industry in delivering 64-bit software by over 24 months.

Compiled-code improvements have been remarkable. In 1992, I could read the code generated by some of our compilers and redline three out of every four instructions as unneeded unneeded unneeded unneeded. A year ago, I could read compiled code and redline one instruction out of every two as unneeded unneeded. Today, I am hard-pressed to redline even 15 percent of the instructions as unneeded.

Moving beyond the installed base, migration efforts are now focused on bringing in new customers. In addition to VAX and MIPS binary translation, the SPARC-to-Alpha binary translation product is available. Code from x86 PC platforms runs emulated on all Alpha operating systems. A technology demonstration of x86-to-Alpha binary translation has been given at trade shows.

The growing maturity and sophistication of the Alpha products have in turn led to accelerated sales growth. Over 100,000 Alpha systems worth over $3.5 billion (hardware, software, and service) have been shipped, and the ship rate has increased 66 percent in the past year alone. In its first three years, Alpha is off to a much faster start than other RISC architectures, such as HP-PA, in their first three years. Buying patterns have shifted from try-one-out to buy-a-fleet-to-run-the-business.

In three short years, Alpha computers have become established as the fastest in the industry--the yardstick by which others measure computer performance. Competitors have shortened their development cycles and aggressively increased their clock rates. Every single company that described Alpha features as unnecessary in 1992 is now rushing to bring its own 64-bit and relaxed read-write order SMP implementations to market. Alpha has grown from "niche design" to "industry yardstick" in a single generation.

Digital has invested over $1 billion in the Alpha development. Literally thousands of people have brought a paper design to life. Bleeding-edge and brute-force chip technology has turned into practical engineering, with a balance of sophistication and everyday care: race cars to express-delivery trucks.

Alpha is evolving much as the architects originally envisioned. I believe Peter Conklin, who led the Alpha Program Office and to whom this issue is dedicated in memoriam, would want to also dedicate this issue to all the brilliant and hard-working people who have made it a reality. My thanks and admiration to each of you.

So how are we doing? After reading this issue, I think you will agree "Quite well, thank you."

Intel486 is a trademark of Intel Corporation. MIPS R4000 is a trademark of MIPS Computer Systems, Inc. SPARC is a registered trademark of SPARC International, Inc. Windows NT is a trademark of Microsoft Corporation.

Skip page footer
printable version
privacy statement using this site means you accept its terms © 1994-2002 hewlett-packard company
End of page footer