1. 1 Introduction Background: CS 3810 or equivalent, based on Hennessy and Patterson’s Computer Organization and Design Text for CS/EE 5810/6810: Hennessy and Patterson’s Computer Architecture, A Quantitative Approach, 3 rd Edition Topics  Measuring performance/cost/power  Designing instruction sets  Instruction level parallelism, dynamic and static  Memory hierarchy  Multiprocessors  Storage systems and networks

2. 2 Organizational Issues Office hours, MEB 3124, by appointment TA: Naveen Muralimanohar, email: naveen, office hrs: TBA Special accommodations, add/drop policies (see class webpage http://www.cs.utah.edu/classes/cs6810/ ) Please sign up for the class mailing list (cs6810) Grades:  Two midterms, 25% each  Homework assignments, 50%, you may skip one  No tolerance for cheating

3. 3 Lecture 1: Measuring Performance How do we conclude that System-A is “better” than System-B? Topics: (Sections 1.1, 1.2, 1.5, 1.6)  Metrics for different market segments  Benchmarks to measure performance  Quantitative principles of computer design

4. 4 Microprocessor Performance 15x performance growth can be attributed to architectural innovations

5. 5 Evolution of Computing Mainframes (1960s)  Minicomputers (1970s)  Desktop PCs (1980s)  Handheld devices (1990s) Today, there are three major microprocessor market segments: desktop, servers, embedded  Different applications  Different requirements  Different architectures

6. 6 Characteristics of Each Segment Desktop: Need high performance (graphics performance?) and low price Servers: Need high throughput, availability (revenue losses of $14K-$6.5M per hour of downtime), scalability Embedded: Need low power, low cost, low memory (almost don’t care for performance, except real-time constraints) FeatureDesktopServerEmbedded System price$1000 - $10,000$10,000 - $10,000,000$10 - $100,000 Processor price$100 - $1000$200 - $2000$0.20 - $200 Design issuesPrice-perf, graphicsThruput, availability, scalability Price, power, app-specific performance

7. 7 Benchmark Suites Standard Performance Evaluation Corporation (SPEC)  CPU2000 (integer and floating-point applications for desktops and servers) stresses CPU and memory, little I/O  SPECviewperf and SPECapc for graphics applications  SPECSFS for file systems, SPECWeb for web servers (to stress disk and network I/O) Transaction Processing Council suites measure throughputs for database transactions Five EEMBC benchmark classes for embedded apps: automotive/industrial, consumer, networking, office automation, telecommunications

8. 8 Summarizing Performance Consider 25 programs from a benchmark set – how do we capture the behavior of all 25 programs with a single number? P1 P2 P3 Sys-A 10 8 25 Sys-B 12 9 20 Sys-C 8 8 30  Total (average) execution time  Total (average) weighted execution time  Average of normalized execution times  Geometric mean of normalized execution times

9. 9 Normalized Execution Times Advantage of GM: no reference machine required Disadvantage of GM: does not represent any “real entity” and may not accurately predict performance Disadvantage of AM of normalized: need weights (which may change over time) Advantage: can represent a real workload

10. 10 GM Example Computer-A Computer-B Computer-C P1 1 sec 10 secs 20 secs P2 1000 secs 100 secs 20 secs Conclusion with GMs: (i) A=B (ii) C is ~1.6 times faster For (i) to be true, P1 must occur 100 times for every occurrence of P2 With the above assumption, (ii) is no longer true Hence, GM can lead to inconsistencies

11. 11 Title Bullet