Introduction:
The Computer History Museum is crucial for preserving and appreciating significant technological advancements and innovations. John Mashey emphasizes the importance of ensuring that valuable information and artifacts are not lost or discarded due to rapid technological advancements.

Preserving Historical Significance:
The information age’s rapid pace can result in important information and artifacts being overlooked or discarded because people may not recognize their significance. The Computer History Museum aims to prevent this loss by actively collecting, preserving, and exhibiting artifacts and information related to computing history. The museum’s efforts contribute to the preservation of knowledge and understanding of the evolution of computing technology.

Membership and Support:
John Mashey encourages attendees who are not yet members of the Computer History Museum to become members and support its mission. Memberships help sustain the museum’s operations, enabling it to continue its work in preserving and showcasing computing history.

The Importance of Preserving Computing History:
The Boston Computer Museum aims to capture and preserve the rapid advancements in computing technology over the past 50 years. Computers have become ubiquitous, and their evolution represents a significant period in human history. The museum seeks to collect physical artifacts, documentation, and other materials related to computing history.

Visible Storage and Notable Artifacts:
The visible storage at the museum showcases a diverse collection of computing artifacts, including personal computers, mainframes, and supercomputers. Some notable artifacts include the Neiman Marcus kitchen computer, Cray supercomputers, and the Johnny Ack machine, an early von Neumann machine.

Exhibits and Events:
The museum hosts exhibits and events that highlight various aspects of computing history, such as the development of the integrated circuit and the rise of personal computers. These events bring together experts, enthusiasts, and the general public to share knowledge and experiences related to computing.

New Museum Building:
The museum recently acquired a new building that was previously a Silicon Graphics office building. The building’s unique design and location make it an ideal space for a technology museum. Plans are underway to transform the building into a state-of-the-art museum with exhibits, storage facilities, and an auditorium.

Call for Support:
The museum relies on donations and volunteers to support its mission. Encouragement to preserve computing artifacts and donate them to the museum. Opportunities for individuals and organizations to contribute financially or through volunteer work.

Personal Involvement and Passion:
John Mashey’s involvement in the museum stems from his belief in the importance of preserving computing history. The museum’s dedicated team is passionate about preserving and sharing the legacy of computing technology. Invitation to the audience to join the effort in preserving computing history.

Backgrounds:
John Hennessy and Dave Patterson had different pathways to computer science. Hennessy started with a tic-tac-toe project using relays in high school. Patterson initially pursued mathematics but switched to computer science after taking a programming class.

Path to Graduate School:
Patterson’s interest in programming led him to pursue a Ph.D. Hennessy completed his undergraduate degree in three and a half years. He sought graduate programs that would accept him mid-year to be closer to his future wife.

Research Interests:
Hennessy’s Ph.D. thesis focused on code generation for microprocessors in real-time applications. This topic became popular as microprocessors emerged, and he completed his thesis at an opportune time.

Stanford and Berkeley:

Hennessy’s Journey to Stanford:
Hennessy interviewed at various universities across the country, including Stanford. Don Knuth’s recommendation and the positive impression of Stanford influenced his decision. He accepted an offer from Stanford and moved to California.

Patterson’s Path to Berkeley:
Patterson’s advisor at UCLA helped him secure a job at the university. He later decided to pursue a Ph.D. and joined the graduate program at Berkeley. Patterson’s family circumstances and desire to stay close to home influenced his choice.

Influence of Computer Science:

Problem-Solving and Algorithmic Thinking:
Hennessy and Patterson emphasized the appeal of computer science as a problem-solving discipline. They highlighted the intellectual challenge and reward of algorithmic thinking. They described the transformative experience of taking a first programming course for many students.

Impact on Career Choices:
Both Hennessy and Patterson discovered their passion for computer science through their early experiences. They were drawn to the field’s problem-solving nature and the satisfaction of creating solutions. Their journeys exemplify the influence of computer science in shaping career paths.

The Influence of Wives on Career Choices:
Dave Patterson’s wife encouraged him to pursue a teaching career, leading him to consider Berkeley as his preferred choice for graduate studies. Patterson’s wife’s strong conviction in their eventual move to Berkeley influenced his decision to apply to the university despite the lack of any initial communication from them.

The Berkeley Application Process:
Patterson initially received a standard rejection response from Berkeley’s chair, Elwin Berlekamp, who informed him that he was ranked in the top 10 but not in the top 5. Later, Patterson discovered that this was Berlekamp’s standard response to many applicants.

The Role of Chance and Opportunity:
Patterson’s application caught the attention of a newly hired faculty member in the hardware area at Berkeley, who found it interesting and invited Patterson for an interview. This opportunity led to Patterson’s eventual hiring at Berkeley, a decision that he credits to his wife’s insistence on contacting the university.

The Start of Dave Patterson’s Journey in Computer Architecture:
Patterson had completed his dissertation on micro-programming tools before joining Berkeley. He took a three-month leave of absence to work at DEC on microprogramming tools for the VAX minicomputer. This experience would later connect Patterson with John Mashey, who also played a significant role in the field of computer architecture.

The Complexity of Microprocessors:
Dave Patterson expressed concerns about the complexity of microprocessors and the challenges in building a VAX-like system without microcode bugs.

Microcode Debugging Techniques:
Patterson proposed a chip design with both ROM and RAM components to allow for microcode bug repairs, but his paper on this topic was rejected.

Risk Architecture Exploration:
In 1980, Patterson and Hennessy began a class to explore RISC ideas, including register windows and delayed branches.

John Koch’s Influence:
John Koch, a renowned computer scientist, visited Patterson and Hennessy and inspired them to pursue RISC architecture.

Koch’s Role in Technology Transfer:
Koch’s goal was likely to communicate RISC ideas to people in Silicon Valley, as IBM had difficulty developing microprocessors.

Delayed Branch Concept:
Patterson and Hennessy adopted the term “Delayed Branch” for their RISC processor, while IBM used the name “Branch and Execute.”

Compilation from High-Level Languages:
The transition from microcode compilation to compiling programming languages like C and Pascal simplified the RISC architecture.

VLIW Architecture Connection:
The RISC and VLIW architectures share a common origin in microprogramming research.

Berkeley’s Contribution:
At Berkeley, Patterson and Hennessy conducted courses and built the first RISC processors in a short time frame.

Carlos Akan’s Expertise:
Carlos Akan, a VLSI expert, joined the RISC project and contributed his knowledge of VLSI design.

John Hennessy’s Background:
John Hennessy’s initial focus was on programming languages and compilers, particularly for the Livermore S1 system.

DARPA’s VLSI Program:
DARPA’s VLSI program, led by Bob Kahn, Dwayne Adams, and John Toole, provided funding and support for the RISC project.

MIPS and RISC Origins:
John Hennessy and Dave Patterson started working on the geometry engine that fueled Silicon Graphics. They developed a microcode compiler and gained insights into generating microcode from higher-level specifications. Work on microprogramming the VAX for the MicroVAX-1 project led to the conclusion that microcode was unnecessary overhead.

MIPS Design Principles:
MIPS project aimed to eliminate the semantic gap between hardware and software by directly translating compiler instructions to hardware level. The project was influenced by the notion of a semantic gap and the belief that microcode was the wrong way to fill it. Register allocation algorithms and compiler technology advancements allowed efficient use of registers in the MIPS architecture.

RISC Design Process:
The MIPS project started with a brainstorming class, similar to the RISC project at Berkeley. The team considered the impact of VLSI technology on processor architecture and focused on optimizing hardware for compiler-generated code. The design process involved extensive brainstorming, group collaboration, and iterative refinement.

Challenges and Controversies:
The MIPS and RISC projects faced resistance from the established computer architecture community. Critics viewed the projects as heretical and questioned their practicality. Despite the skepticism, both projects persevered and demonstrated the viability of RISC architectures.

Experimental Implementations:
MIPS and RISC projects emphasized the importance of building experimental systems to validate ideas. The projects demonstrated that complex designs could be implemented with limited resources, highlighting the value of focused experimentation. The IRAM chip, with 125 million transistors, showcased the feasibility of building complex systems through careful design and implementation.

Evolution of RISC Architectures:
Over time, RISC architectures have continued to evolve, addressing issues such as floating-point support, virtual memory, and language-specific optimizations. The success of RISC architectures led to their widespread adoption in various applications, including desktop computers, mobile devices, and embedded systems.

Lessons Learned:
The MIPS and RISC projects highlighted the importance of focusing on key innovations and making compromises in non-essential areas. The projects demonstrated the value of experimental academic projects in driving innovation and pushing the boundaries of computer architecture. The projects emphasized the need for perseverance in the face of skepticism and the ability to see the potential in revolutionary ideas.

The Birth of MIPS:
John Hennessy and John Massouras, inspired by Gordon Bell’s encouragement, decided to establish MIPS to commercialize their RISC research. They approached investors on Sand Hill Road with a business plan centered on the transformative potential of their technology. Mayfield provided seed funding, enabling the formation of a team and the development of a comprehensive business plan. Initial debates among the founders focused on structuring the company, leading to extensive discussions and planning.

Berkeley Unix and the Connection to Spark:
Berkeley Unix, developed by Ken Thompson and Bill Joy, fostered a connection between Berkeley and Unix. The VACs, chosen by Berkeley instead of PDP-10s, required an operating system, prompting Bill Joy to port Berkeley Unix to the VACs. Bill Joy’s efforts resulted in the development of Berkeley Unix, which coincided with the RISC project.

The Magic CAD Tools:
John Osterhout created CAD tools specifically for RISC chips, known as “magic CAD tools.” These tools proved remarkably durable and are still being used 15 years later.

The Log-Structured File System:
Ideas generated during the RISC project led to the development of a log-structured file system. This innovation earned a Ph.D. for a Berkeley researcher and later became the basis for a successful Stanford professorship.

The Producer-Consumer Relationship between Stanford and Berkeley:
John Hennessy highlights the complex producer-consumer relationship between Stanford and Berkeley. This relationship has resulted in a continuous exchange of ideas and innovations between the two institutions.

Early Sun Days:
Bill Joy was impressed with Manolis Katavinos’ dissertation and decided to start Sun Microsystems in 1981. Dave Patterson was brought in as a consultant on the SPARC project, which was initially codenamed Sunrise. Robert Gardner was recruited from Park to join the SPARC team.

Business Model Challenges:
MIPS visited Sun to propose adopting their instruction set architecture (ISA), but Sun declined due to concerns about the business plan and cost. Sun aimed to sell boards instead of just chips, which was a different approach from other companies. The fabless semiconductor company model did not exist at the time, and Intel emphasized the need for semiconductor companies to own fabs.

Sun’s First Microprocessor:
Sun built its first microprocessor, the Gatorade chip, as a gate array with about a dozen people. The Gatorade chip outperformed Intel and DEC’s VAX in terms of speed, despite being a relatively simple implementation. Sun’s aggressive sales force helped push the Gatorade chip and its advantages in the market.

John Hennessy’s Involvement:
John Hennessy spent a year at Sun as a consultant before returning to academia as planned. His departure caused some concerns among venture firms financing the company due to his role as a key carrier of the technology.

Consulting and Industry Interactions:
Dave Patterson and John Hennessy shared their experiences as academics with extensive industry interactions. They discussed the benefits and challenges of such interactions, including the need to balance academic and industry commitments.

Importance of Industry Experience for Computer Scientists:
Working in industry can bridge the gap between theoretical concepts and practical applications. Practical experience enhances understanding of technology and real-world problems. Industry experience helps identify genuine problems that matter.

Relevance of Industry Experience for Academia:
Industry exposure improves teaching effectiveness. Practical knowledge enhances the ability to select research topics that are relevant and impactful. Industry connections help identify problems that are different from those being addressed by industry.

Importance of RAID Project and 10-Year Reunions:
RAID project as an example of successful collaboration between academia and industry. 10-year reunions provide an opportunity to assess the long-term impact of research projects. Reunions highlight the positive impact of teaching and research on individuals’ lives and careers.

RISC and RAID:
Randy Katz initiated the RAID project after observing the potential of small PC disks. The aim was to build a large disk system using numerous small disks for enhanced performance. The project encountered reliability issues due to the increased number of disks, leading to the addition of redundant disks. RAID gained popularity due to its reliability rather than solely performance-oriented features.

Shared-Address Multiprocessors:
John Hennessy led the exploration of shared-address multiprocessors to address programming challenges with message-passing machines. The team aimed to scale shared memory machines to larger processor counts while maintaining programmer-friendly programming models. They investigated cache coherency techniques, including directory-based mechanisms, to enable scalability. The Dash project demonstrated the feasibility of a shared-address multiprocessor using Xilinx parts and modifications to the Silicon Graphics design. Dave Patterson advocated for shared address in multiprocessor design, which was a minority view at the time but later proved to be the right approach.

Network of Workstations:
The network of workstations project explored the concept of using workstations as building blocks for computing systems. Inspired by Forrest Baskett’s work on using existing chips to build computer architecture, the project aimed to utilize workstations as fundamental components. The project resulted in the development of cluster technology, enabling the construction of systems with hundreds of computers. The project had practical applications, such as database sorting records and serving as the inspiration for Ink2Me.

The Versatility of the MIPS Processor:
The MIPS processor proved to be surprisingly versatile, capable of handling both LINPACK and internet services. For internet services, failure isolation was prioritized over shared address space.

The Shared Address Space Advantage:
Shared address space enabled memory sharing, which was cost-effective in the past. However, the commoditization of DRAM made cluster computing more viable.

The Shift towards Cluster Computing:
DRAM’s transformation into a commodity made cluster computing more feasible. This shift marked a significant change in computing architecture.

Dave Patterson’s Vision of the Future:
Patterson predicted the eventual convergence of DRAM and microprocessor fabrication lines. This vision inspired his work on integrating processor and memory on a single chip.

The Birth of Processor-Memory Integration:
Patterson and his team embarked on a project to develop a chip combining processor and DRAM technology. The goal was to create a compact, energy-efficient solution suitable for handheld devices and media applications.

Cray on a Chip:
The final product resembled a “cray on a chip,” offering the density and latency of DRAM with the capabilities of a processor. This breakthrough paved the way for more compact and energy-efficient computing devices.

MIPS Coprocessor with Multimedia Extensions:
MIPS coprocessor was extended for multimedia applications, demonstrating its versatility beyond floating-point calculations. The vector coprocessor was designed by Christos Kouzourox as part of his dissertation. Surprisingly, the vector architecture resulted in code size savings due to its inherent efficiency.

Collaboration and Support:
Cray donated vectorizing compilers, enabling the creation of optimized vectorized code. IBM provided the fabrication facility and DRAM technology for the chip, contributing to its manufacturing feasibility. Avanti supplied the necessary CAD tools for chip design.

Academic Administration:
John Hennessy’s journey through various academic administrative roles is discussed. He served as department chair, dean of engineering, and eventually became the president of Stanford University. Hennessy highlights the rewards of academic administration, emphasizing the impact on a large community and the pride in witnessing the collective achievements of faculty and students.

Transition from Research to Administration:
Hennessy contrasts the immediate rewards of working closely with students and research groups in academia with the broader, long-term rewards of academic administration. In administration, the focus shifts from individual interactions to overseeing and facilitating the success of a larger academic community.

Initial Motivation for the Book:
Dave Patterson and John Hennessy were frustrated with the books available for teaching computer architecture. They felt the books did not adequately cover the RISC (Reduced Instruction Set Computer) approach, which was gaining popularity at the time.

Process of Writing the Book:
Patterson and Hennessy began working on the book in 1989, with the goal of completing it before Patterson became chair at Berkeley in 1990. They found a publisher, Morgan Kaufman, who agreed to a penalty clause if the book was not completed on time. The authors worked incredibly hard, meeting three times a week in Palo Alto to write and discuss the material. They produced an alpha edition and then a beta edition, incorporating feedback from colleagues and students.

Challenges of Writing the Book:
Patterson and Hennessy had never written a book together before and had to learn how to collaborate effectively. They faced the challenge of writing a comprehensive and accurate book while also making it accessible to students. The rapidly changing field of computer architecture meant that they had to continually update the book with new information.

Success of the Book:
Computer Architecture: A Quantitative Approach became a widely used textbook in computer science programs around the world. The book has gone through multiple editions, with each edition incorporating new material and updates to reflect the latest developments in the field. The book’s success is attributed to its clear and concise writing style, its comprehensive coverage of the material, and its focus on quantitative analysis.

Background:
John Hennessy and Dave Patterson discussed the creation of their influential textbook “Computer Architecture: A Quantitative Approach” (CAAQA). They aimed to create a comprehensive and accurate book that would present key concepts and underlying principles in a clear and systematic manner.

Collaboration and Teamwork:
Hennessy and Patterson worked together as a team, leveraging their expertise and shared understanding of computer architecture to produce a high-quality textbook. They sought feedback from colleagues and connections in the industry to ensure the material was up-to-date and relevant.

Overcoming Publishing Challenges:
The authors faced challenges with the publisher’s desire for a lengthy review process and a reluctance to start production until all chapters were complete. They persuaded the publisher to adopt a “pipelining” approach, allowing for simultaneous work on different chapters and reducing the overall production time.

Effective Writing and Example Crafting:
Hennessy and Patterson emphasized the importance of crafting clear and complete examples to aid readers’ understanding of complex concepts. They pushed each other to improve the quality of the examples and to ensure that they were comprehensive and insightful.

Time Management and Dedication:
Patterson realized the importance of focusing on completing a project rather than starting multiple tasks simultaneously. They dedicated a significant block of time to work on the book, resulting in a relatively quick turnaround time of 12 months, including publisher delays.

Avoiding the “Second Release Syndrome”:
The authors encountered difficulties when updating the second edition of CAAQA due to the significant amount of rewriting required. They learned the importance of allocating sufficient time for revisions and avoiding a piecemeal approach.

Lessons Learned from Past Mistakes:
Hennessy reflected on the lessons learned from past mistakes in computer architecture, such as separating architecture from implementation and overreliance on complex optimizing compilers. He emphasized the need for a systematic approach to computer design and the importance of considering future applications rather than being constrained by existing workloads.

The Future of Instruction Set Architectures:
Patterson and Hennessy discussed the potential longevity of the x86 instruction set in the desktop market due to binary compatibility constraints. They highlighted the growing popularity of RISC architectures in embedded computing, where engineering factors such as cost, performance, and code size are more critical.

Risk-Taking and Innovation:
Dave Patterson emphasizes the need for a shift from the current focus on excessive transistor count and power consumption to a more balanced approach that prioritizes sensible and innovative solutions.

The Need for New Challenges:
Patterson highlights that the pursuit of faster processors has been a primary goal for 20 years, but with the significant performance gains achieved, it is time to explore new challenges and research directions in computing.

The Potential of Parallelism:
He suggests that the future of computing lies in parallelism, where multiple microprocessors can operate concurrently on a single chip, enabling a fundamentally different approach to processing.

Overcoming Software Compatibility Obstacles:
Patterson emphasizes the need to address software compatibility challenges associated with the transition to massively parallel architectures, as existing software may not be directly compatible with such systems.

The Role of Computer Architecture:
He believes that computer architecture plays a crucial role in addressing these challenges, as it encompasses the design of systems with hundreds or thousands of microprocessors, including the development of programming paradigms and applications tailored for such architectures.

Embracing a Systems-Level Approach:
Patterson advocates for a systems-level approach to computer architecture, where the design of hardware, software, and applications is considered holistically to optimize overall performance and efficiency.

Common Research Issues:
Developing new ideas is the most significant hurdle in microprocessor design. The technology to create the solutions already exists; we just need the ideas. Computer architecture can help solve these problems.

Challenges with Research Groups:
Productivity decreases as research groups grow larger. Designing complex architectures like MIPS requires a small, knowledgeable team. The intellectually complicated organizational design of the RTL level is the most difficult part of processor design.

Recommendations for Successful Research:
Radical ideas are necessary to compete with large research groups. Brainstorming in small groups generates more novel ideas. Multi-faculty projects allow for expertise in various areas. Universities provide a good environment for this type of research due to the relationship between faculty and students.

Historical Perspective:
Each generation faces two or three major challenges. We must look at the big picture and learn from history to solve these challenges.

Arrival of Human-Level AI Within 20-30 Years:
Speaker 02 predicts that within 20-30 years, machines will reach human-level intelligence, posing a significant challenge and crisis for humanity.

Underestimation of AI Development:
John Hennessy highlights that previous predictions of intelligent computers have been inaccurate and by large margins, emphasizing the underestimated difficulty of creating truly intelligent systems.

Technological Breakthroughs:
Hennessy attributes the progress in AI to brute force breakthroughs and Moore’s law, but stresses the need for a deeper understanding of how to build intelligent computers.

Learning and Evolution:
Hennessy points out that human intelligence is the result of millions of years of evolution and learning, and replicating this process in computers is a complex task.

Cognitive Psychology and Brain Understanding:
Hennessy believes that advancements in cognitive psychology and understanding the brain’s functions may take 50 years or more to reach a level where we can create truly intelligent AI systems.

Historical Debate on CISC vs. RISC:
Speaker 01 discusses the CISC (Complex Instruction Set Computer) vs. RISC (Reduced Instruction Set Computer) debate, where RISC gained popularity but CISC survived due to its versatility and adaptability.

CISC vs. RISC: A Historical Perspective:
RISC (Reduced Instruction Set Computing) architectures emerged as a response to the complexity and inefficiency of CISC (Complex Instruction Set Computing) architectures like the x86. CISC architectures, exemplified by the VAX and IAX 432, featured a large and complex instruction set that often resulted in slower performance and higher transistor counts. RISC architectures, on the other hand, aimed to simplify the instruction set and rely on faster technology to achieve better performance and lower transistor counts.

The Demise of RISC Architectures:
Despite the initial promise of RISC architectures, they ultimately failed to gain widespread adoption due to several factors: The x86 architecture, while strange and ugly, was relatively simple and had the advantage of a large installed base. Intel’s investment in market momentum and volume production gave the x86 a significant advantage over RISC architectures. The splintering of the RISC market by various companies led to a lack of standardization and volume, hindering cost-effectiveness.

Tragedy of the Commons in the RISC Market:
The failure of RISC architectures can be attributed in part to the “tragedy of the commons,” where individual companies pursued their own proprietary RISC architectures instead of collaborating on a common standard. This lack of cooperation resulted in a fragmented market, preventing RISC architectures from achieving the necessary volume to drive down costs and gain widespread adoption.

Technology Advancements and the Single-Technology Era:
The widespread adoption of CMOS technology as the dominant semiconductor technology led to a single-technology era, where all processors were built using the same technology. This shift from multiple technologies to a single technology favored volume production and cost-effectiveness, further solidifying the dominance of the x86 architecture.

RISC Ideas Adopted by x86:
Despite the decline of RISC architectures as independent entities, many RISC ideas were eventually adopted by the x86 architecture. Intel’s implementation of micro-instructions allowed the x86 to internally translate complex instructions into simpler RISC-like instructions, improving performance. This adoption of RISC principles enabled the x86 architecture to remain competitive and maintain its dominance in the microprocessor market.