Lecture 23 Course Summary Future Predictions and Your Cal Cultural History Professor David A Patterson Computer Science 252 Fall 1996 DAP F96 1 Final Lecture Review CS 252 follow on courses Research style Discussion on Future of CS E Research Learning about your heritage as Cal students future alumni Course evaluation by HKN Pizza at LaVal s DAP F96 2 Chapter 1 Performance and Cost Amdahl s Law perennial pitfall Make the common case fast Integrated Circuits will continue to dominate computer technology 30M to 100M transistors microprocessor by end of decade Cost vs Price Margins pay the workers of the computer industry For better or worse benchmarks shape a field Interested in learning more on integrated circuits EE 241 Advanced Digital Integrated Circuits CS 250 VLSI Systems Design TA Interested in learning more on performance CS 266 Introduction to Systems Performance DAP F96 3 Chapter 2 Instruction Set Architecture What ISA looks like to pipeline Cray load store machine registers simple instr format RISC Making an ISA that supports pipelined execution VAX Making an ISA that minimizes opcode space easy for compilers many addr modes few reg 80x86 importance of being their first Interested in learning more on compilers and ISA CS 264 5 Advanced Programming Language Design and Optimization CS 294 Reconfigurable Computers Wawzyrnek DAP F96 4 Chapters 3 4 Pipelined Implementation Miracle of Pipelining Bandwidth vs latency Superscalar to break single instruction clock cycle limit Hazards Dependencies as limit HW SW techniques to overcome limits Conditional Branches as one Limit branch prediction Memory system as another limit compiler machine organization try to overcome limits Out of order execution paritally overcome some limits at dramatic complexity increase Sustaining 2X increase 18 months rat race DAP F96 5 Processor Performance Over Time Computer Speed Improvement 200 190 Compounded annually for 8 years computers today are 3 times faster than we would have predicted in early 1980s 180 160 SPECmarks 140 117 120 100 80 Historical rate of micrprocessor speed improvement was 35 year until 1986 60 40 20 4 4 6 85 86 As a result of computer research at universities and research labs starting in 1987 the rate of improvement increased to 55 year 80 65 51 49 36 27 Historical trend projected until 1994 35 year 0 84 87 88 89 Year 90 91 92 93 94 DAP F96 6 Appendix B Vector Processors High level operations work on linear arrays vectors Alternate model much easier for hardware more powerful instructions more predictable memory accesses fewer branches longer pipeline Key terms Chime Convoy Chaining Initiation rate Start up time Vector Length Register Strip mining Stride Gather Scatter Vector Mask Register Interesting metrics R speed inifinite vector N1 2 length 1 2 speed R NV length faster than scalar What of computation is vectorizable For new multimedia apps DAP F96 7 Chapter 5 Memory Hierarchy Many many options for caches 4 Questions where who which write 3 C capacity conflict compulsory As CPUs get faster more time spent in memory hierarchy 150 clock cycles to DRAM x 4 instruction issues potentially 600 instruction issues during miss DRAMs continue amazing capacity advance 4X 3 years since 1970s but small advance in latency Memory hierarchy likely overriding issue in algorithms today do algorithms and data structures of 1960s work with machines of 1990s DAP F96 8 Technology Memory Perspective 10 000X increase since 1970 another 50X by 2001 Compared to other phenomena Computer Memory B i l l i o n s 6 000 World Population 5 000 4 000 3 000 Total US Bank Deposits 2 000 1 000 National Debt 0 1970 1975 1980 1985 1990 Since 1970 10X Banks Debt 2X population DAP F96 9 Chapter 6 Storage I O Bandwidth Latency Reliability Queuing theory RAID performance and reliability Disks growing at 4X 3 years more recently Still get email messages to reduce file storage Fantastic potential of tertiary storage 100s TBs Library on Congress at finger tips Interested in learning more on queueing theory IEOR 161 Ross IEOR 267 Wolff IEOR 268 Interested in learning more on SW storage systems CS 286 Implementation of Data Base Systems DAP F96 10 Chapter 7 Networks Similarities of MPP interconnects LANs WANs Bandwidth vs Latency in communication Switches everywhere possibly even replacing memory busses Exciting Area Internet read about in newspaper everyday Who will win Sun 100 Mbit Ethernet HP 100 Mbit Ethernet Switched 10 Mbit Ethernet ATM Interested in learning more on networks CS 268 Computer Networks DAP F96 11 Chapter 8 Multiprocessors Potential for both performance and reliability Shared uniform memory access vs Shared non uniform memory access vs Message Passing Cache coherency protocols Snooping vs directory Successful today for file servers time sharing databases Will parallel programming become popular for production programs If so need to know 3As Architecture Applications Algorithms Interested in learning more on multiprocessors CS 258 Parallel Computer Architecture E 267 Programming Parallel Computers CS 273 Foundations of Parallel Computation DAP F96 12 CS 252 Projects Many many interesting projects Several students and faculty said they enjoyed poster session and mentioned what great jobs you did Many capable of being turned into published papers if you have the time You have seen the full conference cycle topic selection investigation real deadlines poster session written presentation DAP F96 13 Doing Research Don t follow this Bad Career Advice Invent a Field and Stick to it Let Complexity be Your Guide Never be Proven Wrong Use the Computer Scientific Method Avoid Feedback Publishing Journal Papers IS Technology Transfer Write Many Bad Papers Give Bad Talks DAP F96 14 Role Changes during Project P DAP F96 15 Alternatives to a Bad Career Goal is to have impact Change way people do Computer Science Engineering Evaluation of academic research uses bad benchmarks skews academic behavior Many 3 5 year projects gives more chances for impact Feedback is key seek out value critics Do Real Stuff make sure you are solving some problem that someone cares about Taste is critical in selecting research problems solutions experiments communicating results taste is acquired and improved by feedback Students are the coin of the academic realm DAP F96 16 Impact of Industry on Computer Architecture Research in the Future Will PCs drive out all traditional forms of hardware Given cost of IC Fab line increasing to 1B
View Full Document
Unlocking...