The Pentium Chronicles: The People, Passion, and Politics Behind Intel's Landmark Chips

by Robert P. Colwell

"The Pentium Chronicles" describes the architecture and key decisions that shaped the P6, Intel's most successful chip to date. As author Robert Colwell recognizes, success is about learning from others, and "Chronicles" is filled with stories of ordinary, exceptional people as well as frank assessments of "oops" moments, leaving you with a better understanding of what it takes to create and grow a winning product.

Contents

Foreword, by Wen-Mei Hwu
Preface

1. Introduction
2. The Concept Phase
3. The Refinement Phase
4. The Realization Phase
5. The Production Phase
6. The People Factor
7. Inquiring Minds like Yours

Bibliography
Appendix: Out-of-order, superscalar microarchitecture: A Primer. Plausibility Checking.
Glossary
Index

Genres: History, Business, Nonfiction, Computers, Technology, Computer Science, Technical

208 pages, Paperback

First published May 27, 2005

Reviews

[Amar Pai · Rating 3 out of 5]

This book has some interesting stuff on large scale technical project planning, but I got turned off by the endless parade of white men. It's a product of its time, but I hope times have changed. If they haven't, it's because people like Robert P. Colwell haven't examined their privilege.

I mainly picked this up for the Horror Novel frisson you get from reading him blithely describe implementing speculative execution... I'm like "get out of the house!! spectre and meltdown are right around the corner!!"

[Alejandro Teruel · Rating 4 out of 5]

Robert P. Colwell was one of the chief architects of Intel´s daring and polemical Pentium Pro (P6) chip, whose design ran counter to the prevailing wisdom of the times by developing an out-of-order, superscalar microarchitecture for a CISC, rather than a RISC architecture. In order to appreciate this book you need to understand that first sentence, albeit at a basic level. It is a personal, fascinating and very readable account of the engineering tradeoffs and project management issues for a successful technology project that lasted for four and a half years (1991-1995) and got to employ over 400 design and validation engineers.

The book is interesting from a historical viewpoint and throws light on a key technological moment -in http://www.taoyue.com/books/book-revi... you can find a review well-worth reading emphasizing precisely this perspective. However, I am a software engineer, not a hardware engineer and am particularly interested in agile techniques. So what can an agile, object-oriented software engineer take away in 2015 from a book published ten years ago about hardware development carried out twenty years ago?

Hardware, unlike software, has a very tangible production phase. You cannot expect to incrementally design a little, build a little, test a little, rinse and repeat. You have to try to get as much right before you move into silicon, because once you move into production, product and process changes are horribly expensive. So it makes a lot of sense to rely on a waterfall process model and not surprisingly Intel´s P6 four and a half year project was carried out in a four to six stage waterfall process model (concept phase, refinement phase, realization phase, production phase, enriched by pre-silicon and post-silicon validations) under heavy-weight change management. Yet there is an underlying incrementalist approach in the Intel “flagship” and derivations chip approach that hark back to the wildly successful and archetypical IBM 360 architecture and that lurk at the heart of product family development approaches. There are also fascinating glimpses into more lightweight perspectives and techniques: pre-silicon tests (they remind me of acceptance tests and test-driven development), Intel´s production phase “war rooms” (a direct analogy to crunch mode processes which also look forward to Scrum daily meetings and DevOps principles), adding validators to design teams, design inspections and reviews (Colwell actually mentions pair programming and considers it a continuous design inspection technique), a “health-of -the-model metric” (yet another attempt to track progress -and certainly a more sophisticated version of burn-down charts) and even XP´s insistence on more open workspaces are either derived from, inspired by or reinvent many of Colwell (and IBM´s) workspace ideas . As for design principles, separation of concern principles run rife through the P6 project and even though it was clear that performance was a key requirement (in SE terms, a nonfunctional requirement), as an engineer one immediately empathisizes with the host of other requirements that needed to be met (backward compatibility and legacy code performance, for example) and tradeoffs between conflicting criteria, in the chip case, between performance, schedule, target die size, cost and power dissipation.

There are also some fascinating insights into bug dynamics, the idea of testability (and testability and monitoring hooks which would resurface in the agile software world as different kinds of test doubles), and what is nowadays called incident and problem management (e.g. the importance of registering and distinguishing between non-reproducible bugs (a bug “sighting”) and reproducible bugs, the difference beween workarounds and root cause corrections, plus some key warnings well worth repeating: don´t use validation plan as a project performance metric and don´t use number of bugs as a performance measure.

The book also includes a host of interesting managerial pointers albeit be warned that his anecdotes reach a truly dilbertian intensity...

In short, a very enjoyable book with lots of material which can help the appreciative reader develop a deeper perspective into managing hardware or software projects

[Ushan · Rating 4 out of 5]

Robert P. Colwell was the chief architect of the Intel Pentium Pro processor, released in 1995, which had a new superscalar microarchitecture called P6 with speculative execution, out-of-order completion and a long pipeline. The processor was supposed to be twice as fast as the Pentium on 32-bit loads and 10% faster on 16-bit loads (note that in 1995, the latest consumer operating system was Windows 95, released in the summer, which was not yet fully 32-bit). The actual improvement was far more modest, and on some benchmarks negative; BYTE magazine wrote: "Practically any 100-MHz Pentium-based machine will outrace a 150-MHz P6-based computer when running Windows 3.1 applications." The new microarchitecture did pay for itself in the long run, though, and lasted through Pentium III. In an era of rapidly rising clock rates, the goal was, as the preface puts it, to keep parallelism "under the hood", and present the familiar architecture to the application programmer accustomed to sequential programming; how quaint in today's world of multicore and manycore computing! The book describes in detail how they did it. One thing that made me laugh was arranging the cubicles of the chip designers according to the chip layout: if the floating point unit needs to talk to the reorder buffer, the designers of the former need to talk to the designers of the latter. As could be expected for Intel, the P6 team struggled with the legacy instruction set; XOR EAX,EAX is a common idiom in x86 assembly to set register EAX to 0; the out-of-order execution engine recognizes it as a special case that does not depend on the contents of EAX, unlike XOR EAX,EBX. Colwell laughs at the rivalry between the Windows 95 and the Windows NT teams at Microsoft, which he visited to discuss the future of Intel's processors, but later admits that Intel itself had dysfunctional rivalry between the P6 team in Oregon and the Itanium team in Santa Clara. Itanium is one of the costliest failures in the history of engineering; Colwell says that he told the VP in charge of it, whom he does not name, that no company had succeeded doing what he was trying to do with the Itanium, but the VP went ahead anyway - and as we all know, Intel did not succeed either.

[Maynard Handley · Rating 4 out of 5]

To enjoy this book you have to know what it is and what it is not.

(a) It is NOT a book about the technology of Intel or the PPro. If you know even the most basic aspects of OoO processing, speculative execution, and register renaming, then you know everything technical you can learn from this book.

(b) It IS ostensibly a book about how to manager engineers: what motivates them, what demotivates them, how to plan schedules, how to interact with executives. This book is very valuable if you do manage engineers, and is worth reading if you are an engineer so as to understand what your management worries about and is trying to achieve.

(c) There is a second book hidden inside this management treatise, and it is this hidden book that is by far the most entertaining. And this hidden book is an utter hatchet job on Intel's entire top-level executives, marketing operation, and general culture. Colwell is far too nice to come out and say anything explicitly, but reading between the lines it is perfectly clear that he thinks that it was basically an amazing accident that he and his team were able to put together the PPro, and that the natural Intel DNA reasserted itself right afterwards in the crazy designs of the Pentium 4 (designed by marketing purely on the basis of moar moar moar GHz, with zero concern for either power or any other aspects of how to improve CPU performance) and Itanium (designed by a company so high on hubris that it imagined it could boil the ocean, and do so easily).

It's quite clear in the 10+ years since he wrote that he was completely correct in his technical analyses of both these projects (something that was not nearly as obvious at the time of writing), AND, more significantly, that he was correct in his analysis of Intel culture. The specific errors of the P4 and Itanium have been corrected, but the flawed reasoning that led to them, the misunderstanding of computing, Intel's place in it, and Intel's strengths and weaknesses have not, IMHO, been corrected. Hence Intel's loss of the mobile (and now IoT) markets, and its apparent scrambling and lack of direction in the face of a substantially improved AMD, serious process contention from the foundries, and the chance (not yet clear how it will turn out) of real competition from ARM on the "server" front (where server is broadly understood as everything from appliances to various roles within a data warehouse to various types of specialized HPC).

Certainly if you want to understand how Intel lost mobile, even though this book covers the ten years before that happened, it's the place to start.

[CD]

On the 'to investigate' list to delve a bit deeper in the background of how this book came to be published. There's is maybe questionable history or missing precedents within the industry that led to where portions of this story 'begins'.

[Chris · Rating 4 out of 5]

Somewhat repetitive. More about business and managing than electronics. (I'd been looking for the later.)

[Tang Jie · Rating 5 out of 5]

Very nice book about how to implement new cpu
architect and working flow

[Jud · Rating 2 out of 5]

Interesting to hear some of the tales behind the P6 architecture and the internal politics of the Intel project. A lot of the overall content of the book was high level discussions of how to best setup and manage a team for the tremendously difficult task of developing modern microprocessors. I think this book was targeted more at people looking for business/engineering management advice, and I was looking for some a bit more technical, being the geek that I am. This is probably a worthwhile read though if you find yourself in the position of having to manage a group of unruly engineers.

[Jennifer · Rating 3 out of 5]

Great book for somebody new to design/validation semiconductor field. Not very interesting to others. A must read for any new employee at Intel.

[Nick Black · Rating 2 out of 5]

Amazon 2009-03-20.

[Michael · Rating 3 out of 5]

Some nice project management tidbits.

[Carnegie Mellon University Alumni Association]

Robert Colwell (E 1978, E 1985), Author

From the author:
A landmark chip like the P6 or Pentium 4 doesn't just happen. It takes a confluence of brilliant minds, dedication for beyond the ordinary, and management that nurtures the vision while keeping a firm hand on the project tiller. As chief architect of the P6, Robert Colwell offers a unique perspective as he unfolds the saga of a project that ballooned from a few architects to hundreds of engineers, many just out of school. For more than a treatise on project management, The Pentium Chronicles gives the rationale, the personal triumphs, and the humor that characterized the P6 project, an undertaking that broke all technical boundaries by being the first to try an out-of order, speculative super-scalar architecture in a microprocessor. In refreshingly down-to-earth language, organized around a framework "we wish we had known about then," Chronicles describes the architecture and key decisions that shaped the P6, Intel's most successful chip to date. Colwell's inimitable style will have readers laughing out loud at the project team's creative solutions to well-known problems. From architectural planning in a storage room jimmied open with a credit card, to a marketing presentation using shopping carts, he takes readers through events from the projects beginning through its production. As Colwell himself recognizes, success is all about learning from others, and Chronicles is filled with stories of ordinary and exceptional people and frank assessments of "oops" moments, like the infamous FDIV bug. As its subtitle implies, the book looks beyond RTL models and transistors to the Intel culture, often poking fun at corporate policies, like team-building exercises in which engineers ruthlessly shoot down each other's plans. Whatever your level of computing expertise, Chronicles will delight and inform you, leaving you with a better understanding of what it takes to create and grow a winning product.