A Computer Architecture Workshop: - PowerPoint Presentation

1. A Computer Architecture Workshop: Visions for the FutureCelebrating Yale@75September 19th, 2014Trevor MudgeThe University of Michigan, Ann Arbor

2. Visions for the Future?We need to think out-of-the-boxGet back to the basics of computingDo something radical Get rid of the von Neumann straight jacketOur abstractions don’t expose the basic logical fabicI propose the following: Let’s take a leaf from toys—LEGOThere’s a solution that’s been with us a long time....It’s a basic building block that digital systems canbe constructed fromNo –not a NAND gateIt has a higher level of abstraction

3. The Universal Logic Module—ULM

4. Golden Age—Dennard Scaling Worked About every 2 years we got: 2× area 40% increase in frequency 50% reduction in power No change in power density 14

5. Turn of the CenturyCracks appearedFirst crack: frequency scaling became a problemVdd got stuck & power density started to creep upClock frequency could no longer be a selling pointCore count was the new metric Later: Each new tech node no longer showedpower and frequency scalingNode numbers became symbolic names—still followed the 1/√2× sequenceOnly area shrank with new nodes8 64-bit cores

6. What About Cost?Two ViewsIntelEveryone elsecost per transistorreached a minimum at 28 nm

7. Where Does That Leave UsNo “new” technology that looks promisingCheapReliableSupport densityCan’t see much beyond 7 yearsBut where’s the capital investment?From a computer architect’s viewpointFast complex cores have been exploredMultiple cores are being exploreUnlikely to reach large numbers except for data centers / HPCApplication domain specific architectures look promisingGpu led the wayOnly so many applications that warrant specialization

8. Some OpportunitiesMemory—new interesting technologies3D monolithic FLASHVarious NV-memory technologiesAdvanced packaging3D die stackingInterposers Longer time horizon3D monolithic ICs —see FLASHHigh Bandwidth Memory

9. FIN