★ wanayoo — archive 1999 http://www.cpureview.com/art_k7smp_a.htmlNouvelle recherche | Portail wanayoo
CPUReview Logo
Try our discussion Forums!
Feature article:
Choosing the
'Right' PC

Home/News  
Articles  
Reviews  
 


Editorials  
Forum/BBS  
About Linux  
Find LowPrice  

TELL A FRIEND
about us!

Cartoons  
Awards  

AMD  
Cyrix/IBM  
IDT WinChip  
Intel  
Rise  

AboutLinux
Sister Site
  
Buy NerdGear: Tux,Mugs,Shirts

Rumors  
About Us  
Advertising  
Contact Us  
Other Links  
Disclaimer  

All material found on this site is copyrighted by the author, and may not be reprinted, published or used on another site without permission.

All sites are welcome to link to pages in this site, as long as the page is not "framed".

All trademarks mentioned on this page are the property of their respective owners.



 

K7 (Athlon) SMP: What the future holds

by William Henning
Editor, CPUReview
Copyright July 12, 1999
ALL RIGHTS RESERVED

Introduction

Speculation and leaked benchmarks abound about AMD's upcoming Athlon (K7) processor [Editor: We tried to squeeze some information out of AMD in this interview; but they are very tight-lipped] and the latest tidbit of news comes from HotRail Inc.; a San Jose California core-logic start-up.

[Editor: links to HotRail and some references are at the end of this article.]

Their claim to fame is a "point-to-point switched fabric architecture"; basically a gigantic crossbar switch for processors, memory and peripherals that they call "Simultaneous Switched Matrix (tm)" abbreviated as SSM.

In case you are not familiar with such technology it is essentially a "router" for signals between processors and memory. Just like Ethernet switches it tries to implement a point-to-point link between the devices.

HotRail's site describes their SSM device as being appropriate for SMP (Symmetric Multi-Processor) systems with up to eight processors; they document their devices as having up to fourteen ports (for processors, memory and peripherals) with each port providing up to 3.2Gb/sec bandwidth.

The idea is:

  • CPU#1 could retrieve information from memory port #1 while
  • CPU#2 was writing to memory port #3 while
  • an I/O controller is writing to memory port #3 from I/O port 1

And all of these devices can communicate over the SSM without excessive delays introduced due to arbitration (and cache coherency checks).

This is an excellent idea; and similar ideas have been used in super-computers to increase performance.

NEXT: The Problem

Please sign the SMP petition! Click here to open the form in a new window.

Click on links below to find low prices... (Click HERE to search by keywords)
Processors Athlon (K7) K6-3 K6-2 Celeron 'A' 686/M2 Pentium 2/3 Winchip
Motherboards M750i Abit BP6 BH6 LX6 Super 7 Soyo 5EHM Hot 591P SiS530
Video Cards Viper770 Xentor32 TNT2 TNT 128 XPert 128 Voodoo3 V2 G200
Other Goodies ATI-TV SBLive! KDS VS195 DLink 530TX Linux Quake II DVD-RAM
NerdGear2Buy stuffed Tux, Linux/GNU/TCPIP/other T&polo shirts, mugs & more

Go to the home page for news!

Edited by
William Henning
Page design by Web Technologies
Copyright June 5, 1998
All Rights Reserved