Table of Contents
Advertisement
Memory Subsystem
Increasing the data transfer speed has been an important focus area in improving overall system performance. According
to the conventional approach, increasing the data transfer speed between the boards is the key to increasing the system
clock speed. One of the problems related to data transfer has always been the question of how to reduce the deviation of
the clock signals from the data signals (skew). However, a new method (Source-Synchronous method) that involves
sending the clock signal together with the data has been adopted here. This has been found to be very effective in
significantly reducing skew. By this method, data transfer at 540 MHz has been achieved.
Conventional 16-CPU class models require a board on which a crossbar switch is mounted for the connection between
multiple system boards. This requirement causes an increase in memory access latency and an increase in the number of
parts. In designing PRIMEPOWER 650 and 850, the need for this crossbar has been eliminated, and the goal of direct
connection between the system boards has been achieved.
Send side
Send clock
For the address system, a memory controller called a System Controller (SC), where four CPUs and an I/O bridge are
combined into a single unit, controls the cache. On these new models, each SC shares the control and maintains cache
coherency.
01/12/31
PRIMEPOWER 650 and 850 Technical White Paper
Co
nn
Data
ect
Clock
or
Source Synchronous method
Receive Side
WP
RP
Receive Clock
11
Hide quick links:
Advertisement
Table of Contents
