Lecture 07 Multicore Computation Lecture based on notes from John - PowerPoint PPT Presentation

Lecture 07 - Multicore Computation Lecture 07 Multicore Computation Lecture based on notes from John Mellor-Crummey Department of Computer Science Rice University & Jernej Barbic

Lecture 07 - Multicore Computation This was thinking mid-90s.

Lecture 07 - Multicore Computation

Lecture 07 - Multicore Computation Circuit complexity and interconnect delay limit practicality of support structures for larger issue width

Lecture 07 - Multicore Computation

Lecture 07 - Multicore Computation

Lecture 07 - Multicore Computation

Lecture 07 - Multicore Computation

Lecture 07 - Multicore Computation

10 10 Lecture 07 - Multicore Computation

11 11 Lecture 07 - Multicore Computation

12 12 Lecture 07 - Multicore Computation Some important points • Technology alone is not driving push to multi-core – What was state of the art - more issue, superscalar - provides diminishing performance returns b/c of program properties • Still, performance gains possible with scaling • If CCs/instruction performance gains tapped out + scaling performance inhibited (b/c of lower V dd , lower clock rates), where does performance come from?

13 13 Lecture 07 - Multicore Computation Some important points • Performance must come from combination of parallelism + previously ignored HW optimizations – E.g. instead of getting 2x from technology, get 10% from A, 5% from B, etc.

14 14 Lecture 07 - Multicore Computation

15 15 Lecture 07 - Multicore Computation

16 16 Lecture 07 - Multicore Computation

17 17 Lecture 07 - Multicore Computation

18 18 Lecture 07 - Multicore Computation

19 19 Lecture 07 - Multicore Computation

20 20 Lecture 07 - Multicore Computation The cores fit on a single processor socket (also called CMP - chip multiprocessor)

21 21 Lecture 07 - Multicore Computation

22 22 Lecture 07 - Multicore Computation

23 23 Lecture 07 - Multicore Computation Back to case study…

24 24 Lecture 07 - Multicore Computation (standard benchmarks parallelized for comparison)

25 25 Lecture 07 - Multicore Computation

26 26 Lecture 07 - Multicore Computation

27 27 Lecture 07 - Multicore Computation (If CPU time constant, performance comes from parallelism)

28 28 Lecture 07 - Multicore Computation Take Aways

29 29 Lecture 07 - Multicore Computation

30 30 Lecture 07 - Multicore Computation

31 31 Lecture 07 - Multicore Computation

32 32 Lecture 07 - Multicore Computation Multi-core flavors • Cores need not be the same – (If they are, we talk about symmetric core machines) – (If not, asymmetric) • Imagine FPGA + GP processor?

33 33 Lecture 07 - Multicore Computation Other issues: (Amdahl’s Law and Parallelization)

34 34 Lecture 07 - Multicore Computation

35 35 Lecture 07 - Multicore Computation

36 36 Lecture 07 - Multicore Computation

37 37 Lecture 07 - Multicore Computation

38 38 Lecture 07 - Multicore Computation

39 39 Lecture 07 - Multicore Computation

40 40 Lecture 07 - Multicore Computation

41 41 Lecture 07 - Multicore Computation Other issues: Core-to-core communication Must factor in communication costs in processing time too…

42 42 Lecture 07 - Multicore Computation Back to Processor-Memory Wall (still need to feed cores) (Peter Kogge will discuss on Monday) (Not only a problem for multi-core)