Single-Chip Multi-Processors (CMP)

Download Report

Transcript Single-Chip Multi-Processors (CMP)

1
Single-Chip Multi-Processors
(CMP)
ELEC6200-001, Fall 08
PRADEEP DANDAMUDI
2
ELEC6200-001, Fall 08
Microprocessor
Methods To Increase Performance:
• The number of transistors available has a huge effect on the performance
of a processor.
• More transistors also allow for a technology called pipelining.
• Parallelism
3
ELEC6200-001, Fall 08
Parallelism in Microprocessors
• Pipelining is most prevalent
▫
▫
▫
▫
▫
▫
Used in everything
Even microcontrollers
Decreases cycle time
Allows up to 1 instruction per cycle (IPC)
No programming changes
Some Pentium 4s have more than 30 stages!
• Parallelism classifications:
Instruction level
Loop level
Thread level - Future trend
Process level - Future trend
4
ELEC6200-001, Fall 08
Instruction Level Parallelism (ILP)
• Competing technology - Superscalar
• Executing multiple instruction in the same clock cycle.
• Dynamic scheduling-ability to execute out of program order.
• Single processor
• Replace ALU with multiple functional units
• Dispatch several instructions at once
5
ELEC6200-001, Fall 08
Superscalar pipeline
6
ELEC6200-001, Fall 08
Competing technologies
• Simultaneous Multi Threading
▫ Simultaneous Multi threading architecture is similar to that of the
superscalar.
▫ SMT processors support wide superscalar processors with hardware, to
execute instructions from multiple thread concurrently.
• Out-of-Order Execution
▫ Where instructions execute in any order that does not violate data
dependencies.
▫ Note that this technique is independent of both pipelining and
superscalar
7
ELEC6200-001, Fall 08
Centralized architecture
• Disadvantages of centralized architectures such as SMT and
Superscalars are:
- Area increases quadratically with core’s complexity.
- Increase in cycle time – interconnect delays. Delay with wires
dominate delay of critical path of CPU. Possible to make simpler
clusters, but results in deeper pipeline and increase in branch
misprediction penalty.
- Design verification cost high, due to complexity and single
processor
- Large demand on memory system.
8
ELEC6200-001, Fall 08
Why Multiprocessor Systems?
• Single-core microprocessor performance increases are beginning to
slow [1] due to:
▫ Increasing power consumption (>100 W)
▫ Increasing heat dissipation
▫ Diminishing performance gains from ILP & TLP
• As a result manufactures are turning to a multi-core microprocessor
approach
▫ Multiple smaller energy efficient processing cores are integrated
onto a single chip
▫ Improves overall performance by performing more work
concurrently
▫ The latencies associated with chip-to-chip communication
disappear, Shared data structures are much less of a problem.
9
ELEC6200-001, Fall 08
Case for single chip multiprocessors
• Advances in the field of integrated chip processing.
- Gate density
(More transistors per chip)
- Cost of wires
• Large uniprocessors are no longer scaling in performance, because it
is only possible to extract a limited amount of parallelism from a
typical instruction stream using conventional superscalar
instruction issue techniques.
10
ELEC6200-001, Fall 08
CMP Architectures
• Two general types of multi-core or chip multiprocessor (CMP)
architectures
▫ Homogeneous CMPs – all processing elements (PEs) are
the same
▫ Heterogeneous CMPs – comprised of different PEs
• Homogenous dual-core processors for PCs are now available
from all major manufactures
• Heterogeneous CMPs are available in the form of
multiprocessor systems-on-chips (MPSoCs)
11
ELEC6200-001, Fall 08
Single chip Multiprocessor architecture
12
ELEC6200-001, Fall 08
CMP Advantages
• CMPs have several advantages over single processor solutions
▫ Energy and silicon area efficiency
 By Incorporating smaller less complex cores onto a single chip
 Dynamically switching between cores and powering down
unused cores [5]
▫ Increased throughput performance by exploiting parallelism
 Multiple computing resources can take better advantage of
instruction, thread, and process level parallelism
13
ELEC6200-001, Fall 08
14
ELEC6200-001, Fall 08
15
ELEC6200-001, Fall 08
Summary
• The CMP architecture is now the architecture of
choice for semiconductor manufactures
• CMPs are more area and energy efficient than single
processor solutions
• CMPs achieve greater throughput than single
processor solutions as more work can be done
concurrently
• Custom multi-processor systems can now be
designed and simulated from the ground up using
software solutions from several companies
16
ELEC6200-001, Fall 08
References
• http://www.morganclaypool.com/doi/abs/10.2200/S00093ED1V01Y2007
07CAC003
• L Hammond, BA Nayfeh, K Olukotun, “A Single-Chip Multiprocessor,”
IEEE, Sept 1997.
http://occs.ieee.org/presentations/2007/070122_Jenks_ParallelMicroproc
essors.pdf
• Chip Multiprocessor (CMP) Architectures
,web.cecs.pdx.edu/~mperkows/CAPSTONES/DSP1/ELG6163_Burton.ppt
• en.wikipedia.org/wiki/