processor-based systems is one of system-wide balance in which the POWER7 processor
plays an important role.
IBM has used innovative methods to achieve required levels of throughput and bandwidth.
Areas of innovation for the POWER7 processor and POWER7 processor-based systems
include (but are not limited to) the following elements:
On-chip L3 cache implemented in embedded dynamic random access memory (eDRAM)
Cache hierarchy and component innovation
Advances in memory subsystem
Advances in off-chip signalling
The superscalar POWER7 processor design also provides a variety of other capabilities,
including:
Binary compatibility with the prior generation of POWER processors
Support for PowerVM virtualization capabilities, including PowerVM Live Partition Mobility
to and from POWER6 and POWER6+™ processor-based systems
Figure 2-3 shows the POWER7 processor die layout with the major areas identified: eight
POWER7 processor cores, L2 cache, L3 cache and chip power bus Interconnect,
simultaneous multiprocessing (SMP) links, GX++ interface, and two memory controllers.
Figure 2-3 POWER7 processor architecture
2.2.1 POWER7 processor overview
The POWER7 processor chip is fabricated with the IBM 45 nm Silicon-On-Insulator (SOI)
technology using copper interconnects, and implements an on-chip L3 cache using eDRAM.
The POWER7 processor chip is 567 mm
(transistors). Eight processor cores are on the chip, each with 12 execution units, 256 KB of
L2 cache, and access to up to 32 MB of shared on-chip L3 cache.
For memory access, the POWER7 processor includes two DDR3 (Double Data Rate 3)
memory controllers, each with four memory channels. To scale effectively, the POWER7
processor uses a combination of local and global SMP links with high coherency bandwidth
and makes use of the IBM dual-scope broadcast coherence protocol.
40
IBM BladeCenter PS703 and PS704 Technical Overview and Introduction
GX++ Bridge
C1
C1
C1
Core
Core
Core
L2
L2
L2
4MB L3
4MB L3
4MB L3
4MB L3
4MB L3
4MB L3
4MB L3
4MB L3
L2
L2
L2
C1
C1
C1
Core
Core
Core
SMP
2
and is built using 1.2 billion components
C1
Core
L2
L2
C1
Core