Simultaneous Multithreading - IBM Power 720 Overview

2.1.3 Simultaneous multithreading

POWER7+ processors support SMT1, SMT2, and SMT4 modes to enable up to four
instruction threads to execute simultaneously in each POWER7+ processor core. The
processor supports the following instruction thread execution modes:
SMT1: Single instruction execution thread per core
SMT2: Two instruction execution threads per core
SMT4: Four instruction execution threads per core
SMT4 mode enables the POWER7+ processor to maximize the throughput of the processor
core by offering an increase in processor-core efficiency. SMT4 mode is the latest step in an
evolution of multithreading technologies introduced by IBM.
Figure 2-4 shows the evolution of simultaneous multithreading in the industry.
Multi-threading evolution
1995 Single thread out of order
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
Figure 2-4 Evolution of simultaneous multithreading
The various SMT modes offered by the POWER7+ processor allow flexibility, enabling users
to select the threading technology that meets an aggregation of objectives such as
performance, throughput, energy use, and workload enablement.
Intelligent Threads
The POWER7+ processor features Intelligent Threads that can vary based on the workload
demand. The system automatically selects (or the system administrator can manually select)
whether a workload benefits from dedicating as much capability as possible to a single thread
of work, or whether the workload benefits more from having capability spread across two or
four threads of work. With more threads, the POWER7+ processor can deliver more total
capacity as more tasks are accomplished in parallel. With fewer threads, those workloads that
need fast individual tasks can get the performance that they need for maximum benefit.
2004 2-way SMT
Thread 0 Executing
Thread 2 Executing
1997 Hardware multi-thread
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
2010 4-way SMT
FX0
FX1
FP0
FP1
LS0
LS1
BRX
CRL
Thread 1 Executing
Thread 3 Executing
Chapter 2. Architecture and technical overview
No Thread Executing
47