Your SlideShare is downloading. ×
Isca needle a_0610
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Isca needle a_0610

750

Published on

Moving the Needle: Computer Architecture Research in Academe and Industry …

Moving the Needle: Computer Architecture Research in Academe and Industry
By Bill Dally

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
750
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
5
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Moving the NeedleComputer Architecture Research in Academe and Industry
    Bill Dally
    Chief Scientist & Sr. VP of Research, NVIDIA
    Bell Professor of Engineering, Stanford University
  • 2. Outline
    The Research Funnel
    Most ideas fail
    Those that succeed take 5-10 years
    The Research Formula
    Constraints
    The Academic Advantage
    The Industrial Advantage
    Startups
    Best practices
  • 3. Goal – Positive Impact on a Product
  • 4. The Research Funnel
    Technology
    insight
    Concept
    Dev
    Model
    Eval
    Dev
    Applications
  • 5. Most ideas fail
    The ideas that succeed take a long time
    Concept
    Dev
    Model
    Eval
    Dev
  • 6. Most ideas fail
    The ideas that succeed take a long time
    Concept
    Dev
    Model
    Eval
    Dev
  • 7. Most ideas fail
    So terminate the bad ones quickly
  • 8. Most ideas fail
    So terminate the bad ones quickly
    Be a terminator, not an advocate
  • 9. Dally, “Micro-Optimization of Floating-Point Operations, ASPLOS, 1989, pp 283-289
  • 10.
  • 11. Most ideas fail
    The ideas that succeed take a long time
    Concept
    Dev
    Model
    Eval
    Dev
  • 12. The ideas that succeed take a long time
    So aim research 5-10 years ahead of current practice
  • 13. Current Architecture Practice
  • 14.
  • 15. Aim Here
    5-10 years
  • 16. Enable this point
    5-10 years
  • 17. Timeline for some ideas
  • 18. The Performance Equation
  • 19. The Research Formula
  • 20. Reward
    If you are wildly successful, what difference will it make?
  • 21. Effort
    Learn as much as possible with as little work as possible
  • 22. Effort
    Do the minimum analysis and experimentation necessary to make a point
  • 23. Real and Artificial Constraints
  • 24. Constraining Infrastructure
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 25. Constraining Infrastructure
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 26. Constraining Infrastructure
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 27. The contribution is insight
    Not novelty
    Not numbers
  • 28. Research is a
    hunt for insight
    Need to get off the beaten path to find new insights
  • 29. Road-Kill Research
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 30.
  • 31. Looking here for lost keys
  • 32. Lost keys here
    Looking here
  • 33. The Academic Advantage
  • 34. The Academic Advantage
    Freedom
  • 35. The Academic Advantage
    Freedom from artificial constraints
    Freedom to fail (take risks)
  • 36. Academic research matched for early stages of the funnel
    Concept
    Dev
    Model
    Eval
    Dev
  • 37. Example: ELM
    An Ensemble
    Many Ensembles and memory tiles on a die
    37
  • 38. Example: ELM
    Balfour et al., "An Energy-Efficient Processor Architecture for Embedded Systems" CAL, Jan. 2008, pp 29-32.
  • 39. ELM Infrastructure
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
    Changed for ELM
  • 40. The Industrial Advantage
    Resources and Experience
  • 41. The Industrial Advantage
    Resources to carry out detailed studies
    Experience to address commercial constraints
  • 42. The ideal partnership:
    Academic research 5-10 years out, focused on industry problems
    Transfer insight to industrial research to refine into product
    Concept
    Dev
    Model
    Eval
    Dev
  • 43. What transfers is insight
    Not academic design
    Not performance numbers
  • 44. What transfers is insight
    And its transferred by people
    Not papers
  • 45. Academic
    Concept
    Analysis
    Simulation
    Prototype
    Refine Concept
    Detailed Design
    Industrial
  • 46. Industrial
    Academic
    Gap
    Concept
    Analysis
    Simulation
    Prototype
    Refine Concept
    Detailed Design
    Impact
    Paper
  • 47. Example: Cray T3D and T3E
  • 48. J-Machine
    • MIT 1987-1992
    • 49. 3-D network
    • 50. Global address space
    • 51. Fast messaging and synchronization
    • 52. Support for many models of computation
  • Cray T3D
    • Started working with Cray in 1989
    • 53. Project started early 1990
    • 54. First ship in mid 1992
    • 55. From J-Machine
    • 56. Network
    • 57. Fast communication/sync
    • 58. Global address space
    • 59. For reality
    • 60. Alpha processors
    • 61. MECL gate arrays
    • 62. Robust software stack
  • Best Practices for Academics
    • Long-term perspective (5-10 years)
    • 63. Know your customer and their long-term issues
    • 64. Look at tomorrow’s applications, not yesterdays
    • 65. Maximize reward, minimize effort
    • 66. Estimate maximum impact – terminate…
    • 67. Minimal analysis and experiment to make the point
    • 68. Exploit your freedom
    • 69. Don’t be limited by exiting tools, benchmarks, ISAs, …
    • 70. Carry result to impact
    • 71. Build relationships with industry
    Benchmarks
    Binaries
    Compiler
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 72. Best Practices for Industry
    • Leverage academic research
    • 73. Build partnerships
    • 74. Articulate long-term research issues
    • 75. Be open-minded
    • 76. Minimize artificial constraints
    • 77. Carry concepts across “the gap”
    • 78. Open infrastructure
  • A Partnership
    Filtered, De-risked Concepts
    Academe
    Industry
    Future issues
    Infrastructure
  • 79. The Startup Path
    When you can’t find an appropriate industrial partner, make one.
    STAC, Avici, Velio, SPI
  • 80. Academic
    Concept
    Analysis
    Simulation
    Prototype
    Refine Concept
    Detailed Design
    Startup
  • 81. Startup Pros/Cons
    Pros
    • Don’t have to convince existing company to change course (until exit)
    Cons
    • Have to convince investors (repeatedly)
    • 82. Have to build a whole company, not just a development team
    • 83. Finance, sales, marketing, …
    • 84. Limited resources
    • 85. Impatient capital
  • Example: SPI
  • 86. Much easier to license technology to an existing company
  • 87. Starting a company to bring a new semiconductor product to market costs $30M (to cash flow positive)
    If it’s a programmable processor, its $70M
    Investors want a 10x ROI
    Need to see a $700M exit to justify a new processor company
  • 88. The future of computer architecture
  • 89. The future of computer architecture
    • NOW is an ideal time for research to move the needle
    • 90. Computers are drastically changing
    • 91. Pervasive parallelism
    • 92. Energy limited
    • 93. Bandwidth constrained
    • 94. Opportunity to set the MSB of future computers in the next few years
    • 95. Requires changing the whole stack
    • 96. Requires industry-academe partnership
  • Energy-Efficient ArchitectureAbstracting Locality
    20mm
    L3
    7pJ
    2000pJ
    Net
    50pJ
    500pJ
    L2
    Net
    2000pJ
    L1
    L1
    L1
    L1
    P
    P
    P
    P
  • 97. Solution involves many levels of the “stack”
    Application
    Algorithm
    Prog. System
    Compiler
    ISA
    uArch
    Too constrained to innovate within one layer
    Design
    Circuits
    Process
  • 98. Benchmarks
    Binaries
    Compiler
    Academe
    Industry
    Simulator
    uArch Idea
    ISA
    Other
    uArch
  • 99. Moving the NeedleComputer Architecture Research in Academe and Industry
    Bill Dally
    Chief Scientist & Sr. VP of Research, NVIDIA
    Bell Professor of Engineering, Stanford University

×