ut_chiou_arch_10 - Real Real-World Architecture G Glenn...

Info iconThis preview shows pages 1–6. Sign up to view the full content.

View Full Document Right Arrow Icon
eal orld Architecture Real-World Architecture Glenn Henry G Glenn Henry Centaur Technology 1
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
The Real-World Challenge ± Make the right choices among the many conflicting objectives At least 6 major areas of optimization for any product Each objective affects the other 5, sometimes drastically Different products have different priorities Performance (Always important, x IPC GHz (Important by itself, esp. in x86) but not the only important thing) Power Consumption Function ery important in mportant in x86) Cost Schedule (Very important in some markets) (Important in x86) (Time = competitiveness) (Very important in some markets)
Background image of page 2
The Real-World Challenge ± These choices significantly affected by implementation details Resource & skill constraints Tradeoffs with other product developments Patent issues Technology constraints eg wires are getting relatively slower than transistors temperature inversion effect re transistors & wires transistor speed improvement is slowing akage is increasing leakage is increasing minimum voltage has stopped lowering models are getting less accurate (esp. at low V) ± The result… It is impossible to make “right” architecture tradeoffs “up front” We spend more time on “architecture” during development then we do “up front” during architecture/planning phase
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Why So Hard To Get It Right? ± Nauseous ISA details (especially x86!) Usually ignored in the literature ± Critical IPC differences are in the details (esp. x86) Requires cycle-accurate modeling with real target code ± Aggressive GHz goals Must use custom design approaches Î reduces flexibility Optimum architecture must be detuned to meet cycle time Need data for tradeoffs before implementation exists ± Catch 22 physical design Wiring delays becoming the dominant factor More IPC Î more wire Î bigger Î slower GHz Î need more IPC ± Difficulty & accuracy of IPC prediction ± Difficulty & accuracy of GHz prediction ± Difficulty & accuracy of power consumption prediction
Background image of page 4
Performance Design Challenges ± How to define “performance”? GHz vs. IPC, CPU vs. system, MP vs. UP, thruput vs. latency, benchmarks vs. architectural, quantification formulae, etc. ± What benchmarks to use? Spec not very important or representative in x86 world Over 200 x86 “benchmarks” reported, wide variance in behaviors All are important Î each customer has their favorite ± How predict performance on new design? Analytic model vs. “cycle-accurate” model vs. real RTL vs. ? What about the system (non-processor) components?
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 6
This is the end of the preview. Sign up to access the rest of the document.

This note was uploaded on 09/23/2010 for the course EE 360n taught by Professor Staff during the Spring '08 term at University of Texas at Austin.

Page1 / 18

ut_chiou_arch_10 - Real Real-World Architecture G Glenn...

This preview shows document pages 1 - 6. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online