1 / 8

If You Like Money, General-Purpose I s for You

If You Like Money, General-Purpose I s for You. Chris Hughes Parallel Computing Lab Intel. Hardware Specialization. Affirmative claim: general-purpose computing = dead end “We can’t turn on all our transistors anyway” “Diminishing gains from general-purpose techniques”

tirzah
Télécharger la présentation

If You Like Money, General-Purpose I s for You

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. If You Like Money,General-Purpose Is for You Chris Hughes Parallel Computing Lab Intel

  2. Hardware Specialization • Affirmative claim: general-purpose computing = dead end • “We can’t turn on all our transistors anyway” • “Diminishing gains from general-purpose techniques” • “Heterogeneous systems only way to achieve big efficiency gains” • Yes, the sky is falling

  3. Reality • Heterogeneous systems nightmare for software developers • Also, gains limited from data/control transfer • Further, we can capture most gains of specialized HW w/o punishing app developers • How do I know this? • We do it all the time • That’s how we make money

  4. It’s All About Money • Greed is good • App developers: New functionality = • Re-codingdoesn’t make money • Existing code must improve automatically • System vendors: • Improve existing apps that drive sales = • Make it easy to add new functionality =

  5. How To Make Money • Don’t mess with the programming model • Don’t sacrifice performance on existing apps

  6. Specialization Breaks The Rules • Specialization almost always does one or both • Expose that work or data migrates • Performance on legacy code degrades • Very hard to implement algorithmic advances • Etc.

  7. Good News! • Industry likes making money • Thus, we already follow above rules AND capture most benefits of specialization

  8. Specialization’s Real Role • Research into specialization helps drive evolution of general-purpose systems • Proof by example: • Floating-point instructions, encryption instructions, wider SIMD & introduction of many-core • Trying to sell specialized systems means: • (1) You hate money • (2) You hate programmers

More Related