ChipCenter Questlink
SEARCH CHIPCENTER
Search Type:
Search for:




Knowledge Centers
Product Reviews
Data Sheets
Guides & Experts
News
International
Ask Us
Circuit Cellar Online
App Notes
NetSeminars
Careers
Resources
FAQ
EE Times Network
Electronics Group Sites

EE Expert David Gilbert
SpacerMicroP and MicroC Architecture

Click here to go to the MicroP & MicroC Archive Click here to go to David Gilbert's Main PageClick here to go to the Guides and Experts Main Page

A Chip Off the Old Block
by David Gilbert

PowerPC 970

Introduction

IBM announced plans on October 15, 2002 at the Microprocessor Forum to release the first in a new family of 64-bit PowerPC chips based on the POWER4 architecture. The PowerPC 970 will feature a native 32-bit mode that is capable of running 32-bit application code with no modification, making it a sure bet for early adoption. This will also be necessary for a smooth transition from pure 32-bit environments to 64-bit environments. Let's see what we have to look forward to from the PowerPC 970.

PowerPC 970 vs. POWER4

The Difference?

The POWER4 is a dual-core chip, so designers remapped one of the POWER4 cores to use 0.13 µm CMOS Silicon-on-Insulator (SOI) technology, and increased the L2 cache size to 512K and the clock speed to 1.8 GHz. The result is a smaller die that uses less power, and yet still has very impressive throughput capabilities. The L1 caches are 64K for data and 32K for instructions, and these feed two fixed-point arithmetic units, two floating-point units, two load/store units, a highly complex branch unit, and the AltiVec-capable vector unit.

What Can It Do?

In a single cycle, the PowerPC 970 is capable of issuing up to 8 instructions while dispatching as many as 5. This is accomplished through an I/O interface bus that operates at up to a 900 MHz clock speed, moving 6.4 Gbits/s of data. The PowerPC 970 also supports Symmetric Multi-Processing (SMP) for applications that can benefit from it. The integrated Memory Management Unit (MMU) is capable of 42-bit physical addressing and 64-bit virtual addressing, and the branch unit can handle up to two branches per cycle. All told, the PowerPC 970's peak throughput is 7.2 GFLOPs per second when executing scalar instructions, or 14.4 GFLOPs per second when executing SIMD instructions.

The 576-pin CBGA chip consumes 42 W at 1.3 V at a clock speed of 1.8 GHz. This is a bit "hot" compared to embedded designs, but this chip was intended for desktop computers and workstation/server equipment.

Conclusion

The PowerPC 970 will be sampling in the second quarter of 2003, with production ramping up sometime thereafter in the second half of the year. By that time, Intel's chips will have a substantial lead in sheer clock speed, but it is doubtful that software optimization will "catch up" to the Pentium 4 processor soon enough. This leaves the PowerPC 970 processor in an excellent position to become the number one performer in its market segment.

Additional Reading

 

Microprocessor & Microcontroller Architectures Archive

Guides and Experts   Analog Avenue   EDA Tools   PLD   DSP   EDA   Embedded Systems   Power   Test

Click here to get your listing up.

Copyright © 2003 ChipCenter-QuestLink
About ChipCenter-Questlink  Contact Us  Privacy Statement   Advertising Information  FAQ