Multicore Processors

1
Multicore Processors

• Kahraman D. Akdemir
• Department of Electrical and Computer Engineering
• EE505 Project Presentation
• Fall 2005

2
Multicore Processors

• Processor performance
• Moores Law
• Decreasing transistor sizes
• Increasing number of transistors
• Decreasing chip sizes

Higher clock rates More complex Faster
microprocessors
Fig. 1 Moores Law Transistors on a chip double
every 2 years(Gelsinger, 2001)
However Transistors cant shrink forever
Several problems with this approach
Performance increase is slowing down
1990-2000 60 per year 2000-2004 40
per year
SOLUTION MULTICORE PROCESSORS
3
Outline

• Limitations of single core architectures
• Power consumption
• Heat generation
• On chip wires and interconnect latency
• Limited parallelism
• Design time and complexity
• Multicores (CMPs) as a solution
• Structure of multicore processors
• Advantages and improvements
• Task management and parceling
• Multicore examples
• Multicore related results
• Multicore VS single core
• Multicore VS Superscalar SMT
• Impact of number of cores and threads

4
1- Limitations of single core architectures

• Power consumption
• Smaller gates
• Higher clock rates
• 2-3 power increase per 1 performance increase
• Leakage power scales 5X in each technology node

• Heat generation
• Extensive heat rejection
• of multiple kilowatts is
• undesirable from a thermal
• perspective

Fig. 2 Lead microprocessor power increases beyond
expected(Gelsinger, 2001)
5
1- Limitations of single core architectures

• On-chip wires and interconnect latency
• Faster gates and larger chips
• On-chip wire delay increases
• Interconnect delays become more critical
• Limited parallelism
• Finite amount of ILP
• Interdependent instructions
• Design time and complexity
• Hard to design
• Large number of transistors
• Complex methods to increase ILP

6
2- Multicore processors as a solution
a) Structure of multicore processors
Fig. 3 Stanford HYDRA Multicore
structure(Hammond, 2000)
7
2- Multicore processors as a solution

• Advantages and improvements
• Performance increase by means of parallel
  execution
• Power and Energy efficient cores (Dynamic power
  coordination)
• Minimized wire lengths and interconnect latencies
• Both ILP and thread level parallelism
• Reduced design time and complexity
• Task management and parceling
• Usually operating system distributes the
  application to the cores
• Can be done according to resource requirements
• Multicore examples
• AMD Opteron Enterprise, Athlon64
• Intel Pentium Extreme edition, Pentium D
• IBM Power 4, Power 5 and Sun Niagara

8
3- Multicore related results
a) Multicore VS single core
Fig. 4 Multicore VS Single core(Geer, 2005)
9
3- Multicore related results
b) Multicore VS Superscalar SMT(Simultanoues
multithreading)
Fig. 5 Multicore VS Superscalar VS SMT(Hammond,
1997)
10
3- Multicore related results
c) Impact of number of cores and threads
Fig. 6 Impact of number of cores and
threads(Chaudry, 2005)
11
Conclusion

• To Sum Up...
• Improvement needs
• Multithreaded and parallel software/applications
• Improved packaging for high number of I/O pins
• Cash size VS possible number of cores
• QUESTIONS?