Download presentation
Presentation is loading. Please wait.
Published byKristopher Berry Modified over 9 years ago
1
Computer Architecture Challenges Shriniwas Gadage
2
Computer Architecture Computer architects design computer systems – Processors: Intel Pentium 4, IBM PowerPC – Also: memory systems, interconnections, ??? slide 2 Pentium cache memory (DRAM) bridge to I/O disk ethernet card
3
Computer Architecture 3 If the Intel Pentium4 has a faster clock speed than the IBM Power4, does it execute your programs faster? Completing instruction Clock tick Case 1: Case 2: Time
4
Microarchitecture Micro-Architects design processors Goals for processors: – Faster!!!! – Higher bandwidth communication with memory system – Backward-compatible with previous models How do we make processors faster? – Faster clocks (>2 GHz) – Do more work (execute instructions) at same time slide 4
5
5 A Typical Microprocessor Branch Predictor Decode & Rename Issue Logic ALU L2 Cache L1 Instr Cache L1 Data Cache Register File
6
Multiprocessor Architecture Computer systems with multiple processors slide 6 Node Interconnection Network Interconnection Network Pentium cache memory (DRAM) bridge to I/O disk ethernet card Node
7
Multiprocessor Architecture How do we make processors work together? – Exploit parallelism in applications Example: web server – Each processor handles different requests – Processors communicate occasionally to synch up Some challenges: – Interconnection network design – Protocols for communicating and sharing data – Scalability – Reliability slide 7
8
Computer Architecture To a large extent, computer architecture determines: the number of instructions used to execute a program the time each instruction takes to execute the idle cycles when no work gets done the number of instructions that can execute in parallel 8
9
Technology Trends We design an architecture for a given technology Technology parameters: – Number of transistors on a chip – Transistor speed – Amount of memory – Memory speed – Bandwidth between components – Power usage – Applications to be run on system All of these change dramatically over time! slide 9 time parameter
10
Technology Trends Performance was the ultimate metric Transistors were a limiting factor As on-chip transistors became available in the 90s, more functionality and complex circuitry was added to boost performance
11
Technology Trends – A Few Examples Number of transistors – Doubles every 18 months (Moore’s Law) Memory size – 1992: We bought extra 512Kbytes for desktop – 2002: Desktop came with 512 Mbytes – 2012: Desktop comes with 2GB Power usage – Pentium 4 can draw 50 amps of current and burn 50 W Important applications – Word processing, spreadsheets multimedia, web surfing slide 11
12
Technology Trends – Good or Bad? Pessimist: trends make designs obsolete – But now I have to re-think everything I’d already solved! Optimist/Architect: trends offer opportunities – What can we do with a billion transistors? Good design ideas Bad design ideas E.g., It was good idea to scale up processor sizes – But, it now uses too much power and is too complex slide 12
13
What To Do With a Billion Transistors? A.Make the processors bigger slide 13 Proc B.Make more little processors Proc chip Proc
14
Hitting the Wall: Architecture Challenges Single core performance Memory Complexity Power and temperature-efficient designs
15
Hitting the Wall: Architecture Challenges Functionalities in multi-core chips Simplifying the programmer’s task Efficient interconnects Designs tolerant of errors
16
16 Hitting the Power Wall Power is as important a metric today as performance
17
17 The Advent of Multi-Core Chips In the past, performance magically increased by 50% every year In the future, this improvement will be only ~20% every year … unless … the application is multi-threaded! Core Cache bank
18
Interconnects as a Bottleneck In the past, on-chip data transmission on wires cost almost nothing Interconnect speed and power has been improving, but not at the same rate as transistor speeds Hence, relative to computation, communication is much more expensive In the near future, it will take 100 cycles to travel across the chip 50% of chip power can be attributed to interconnects
19
19 Interconnects in Multi-Core Chips A L1 A CPU 3 CPU 1CPU 2 L2 cache L2 control AA A A A L2 control
20
20 Not all Wires are Created Equal B-WiresL-WiresW-WiresPW-Wires Relative latency 1x 0.5x 1.6x 3.2x Relative area 1x 4x 0.5x0.5x Dynamic power (W/m) 2.65 1.46 2.9 0.87 Static Power (W/m) 1.02 0.57 1.160.31
21
21 Data Transfers have Varying Needs Example of a cache coherence transaction: Read exclusive request for a shared block
22
22 Other Interconnect Choices Optical interconnects: speed of light, cost in converting between optical and electrical domains 3D chips: reduces communication distances, low cost for vertical signal transmission, increase in power density
23
23 3D Layouts Cluster (a) Arch-1 (cache-on-cluster)(b) Arch-2 (cluster on cluster)(c) Arch-3 (staggered) Cache bankIntra-die horizontal wireInter-die vertical wire Die 1 Die 0
24
24 Upcoming Architecture Challenges Improving single core performance Functionalities in multi-core chips Simplifying the programmer’s task Efficient interconnects Power and temperature-efficient designs Designs tolerant of errors Clustered architectures: relatively low complexity scalable solution easily handles multiple threads
25
25 Upcoming Architecture Challenges Improving single core performance Functionalities in multi-core chips Simplifying the programmer’s task Efficient interconnects Power and temperature-efficient designs Designs tolerant of errors Heterogeneous perf/power Cores that execute the OS Cores that verify results
26
26 Upcoming Architecture Challenges Improving single core performance Functionalities in multi-core chips Simplifying the programmer’s task Efficient interconnects Power and temperature-efficient designs Designs tolerant of errors Hardware to support transactional memory
27
27 Upcoming Architecture Challenges Improving single core performance Functionalities in multi-core chips Simplifying the programmer’s task Efficient interconnects Power and temperature-efficient designs Designs tolerant of errors Faults are caused by high energy particles that deposit enough charge to toggle bits Variations in conditions may cause a circuit to not produce its result in time
28
Ways to Evaluate New Architectures slide 28 Simulating Modeling Building Tradeoff between three desired features
29
Building Construct a hardware prototype Advantages +Way cool to show off hardware +Runs quickly Disadvantages – Takes long time to build – Expensive – Not flexible Generally too labor intensive for research studies slide 29
30
Modeling Mathematically model the system – Use probabilities and/or queuing models Advantages +Very flexible +Very quick to develop +Runs quickly Disadvantages – Cannot capture effects of system details – Architects are skeptical of models Generally OK for back of the envelope estimates slide 30 mem time = hit time + miss rate*penalty
31
Simulating Write a program that mimics system behavior Advantages +Very flexible +Relatively quick to develop Disadvantages – Runs slowly (e.g., 30,000 times slower than hardware) Method of choice for most architectural research slide 31
32
Simulation Challenges slide 32 Simulator Application System description Performance results Tough problems associated with each arrow!
33
Describing Simulated System How detailed must our simulator be? Model every transistor in the processor? – Would take too long Abstract away details of processor organization? – Could miss important effects of processor features – Could achieve wrong conclusion Need balance – Model in detail only where necessary – E.g., Model memory system in detail, but abstract disks slide 33
34
Why Architects Need Friends Architecture is considered both computer engineering and computer science Architects interact with other areas – Circuit design (Electrical Engineering) – Transmission lines (EE) – Power (EE, Mechanical Engineering) – Compilers (Comp Sci) – Operating systems (CS) – Networking (EE, CS) – Databases (CS) – Queuing theory (CS, EE, Industrial Engineering) slide 34
35
How Architecture Relates to Other Areas slide 35 Computer Architecture Operating Systems, Compilers, Networking Software Circuits, Wires, Network Hardware Application Software Besides these interactions, also global issues! –Power, system verification, performance analysis, etc.
36
How Architecture Relates to Hardware (EE) slide 36 Computer Architecture Operating Systems, Compilers, Networking Software Circuits, Wires, Network Hardware Application Software Architecture should enable efficient hardware design –Avoid huge hardware structures –Avoid cross-chip wires
37
How Architecture Relates to System Software slide 37 Computer Architecture Operating Systems, Compilers, Networking Software Circuits, Wires, Network Hardware Application Software Architecture should support system software –Provide good target for compiler –Support important OS features (such as synchronization)
38
How Architecture Relates to User Software slide 38 Computer Architecture Operating Systems, Compilers, Networking Software Circuits, Wires, Network Hardware Application Software Architecture should efficiently run important apps Intel added MMX hardware to support media apps Sun & IBM design multiprocessors for commercial apps
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.