MSI T - PowerPoint PPT Presentation

1 / 23
About This Presentation
Title:

MSI T

Description:

Durwyn Williams/Channel Manager. Today's Topic: The Game Goes On. AMD Where are you? ... Durwyn Williams/Channel Manager. The 64-bit Question still remains... – PowerPoint PPT presentation

Number of Views:91
Avg rating:3.0/5.0
Slides: 24
Provided by: Kevin5
Category:
Tags: msi | manager

less

Transcript and Presenter's Notes

Title: MSI T


1
Presentation
2
The Game Goes On 2003 Industry Update
  • Anand Lal Shimpi
  • Editor in Chief CEO
  • www.anandtech.com

3
What is AnandTech?
  • Established in April 1997
  • Worlds largest PC Technology Website
  • Approaching 4 million cookied monthly readers
  • 60 million monthly page views
  • In-depth product reviews technology analysis
  • Up-to-date coverage of news tradeshows
  • Catering to the Enthusiast user

4
Todays Topic The Game Goes On
  • AMD Where are you?
  • NVIDIA Becomes a Credible Chipset Manufacturer
  • Intels Newest Chipsets What they Mean
  • Centrino
  • The 64-bit question still remains

5
AMDs Quiet Front
  • Athlon 64 1 Year Later
  • Demonstrated at Spring IDF in 2001 at
  • Still waiting for introduction
  • High clock speeds will be necessary to compete
    with Prescott later this year
  • 2GHz necessary to compete with 3.2 3.4GHz
    Prescott
  • Benefits of larger cache are reduced by presence
    of on-die memory controller (and vice versa)

6
AMDs Quiet Front gets Louder
  • Opteron is Upon us
  • Strange naming system
  • Model xyz
  • x number of processors supported in SMP
  • yz relative performance (dependent on
    frequency, cache size, etc..)
  • Allows AMD to remove focus from clock speed while
    distance the Opteron from desktop model numbers
  • Pro Desktop apps no longer influence server
    model numbers
  • Con The confusion continues
  • Hints of unusually low motherboard availability
    until Summer

7
NVIDIA Now a Credible Threat
  • nForce2 is the best Athlon XP chipset currently
    available
  • Stable
  • Solid Drivers
  • High Performance
  • Finally seeing a threat from VIA (KT400A)

8
Intels New Chipsets
  • Springdale (865) Canterwood (875)
  • Bring 850E-class performance to the mainstream
  • 800MHz FSB Support
  • Key to enabling higher frequency CPUs
  • Bandwidth balanced with 128-bit DDR400 interface
  • Very high pincounts
  • Very nForce2-like ball-layout
  • Serial ATA now mainstream
  • as well as RAID

9
Intels New Chipsets
  • Canterwood is it worth it?
  • Performance data not yet public
  • Basically a hand picked Springdale
  • Will replace the 850E
  • 850E 875 upgrade wont be as dramatic as 845
    865 upgrade
  • Just like 850E, performance will be offered at a
    premium
  • Canterwood first, then Springdale
  • Enable new CPUs first
  • then go mainstream
  • Will support first revisions of Prescott
  • Prescott will eventually transition to new
    Socket
  • Whats next? Tejas 1.2GHz FSB

10
Centrino
  • Redefining the way mobile CPUs are designed
  • e.g. if something is too fast, slow it down
  • Very power efficient cache
  • Power optimized 855 MCH
  • Some very impressive initial offerings
  • IBMs Thinkpad T40/X31 series
  • Designs already under way for Dothan (2MB L2!)

11
The 64-bit Question still remains
  • We know AMDs 64-bit strategy Top to Bottom
  • Opteron on the server side
  • Athlon 64 for desktops
  • Mobile Athlon 64
  • but what about Intel?
  • Itanium for high-end 64-bit market
  • IA-32 processors for everything else
  • Deerfield in between?

12
One possibility Yamhill
  • Similar strategy to x86-64
  • Take very strong IA-32 core and make minimal
    changes to add 64-bit support
  • Most likely wouldnt use AMDs branded x86-64
  • Likelihood?
  • If Athlon 64 is successful then very likely
  • Otherwise

13
Something more far fetched
  • High clock speed Itanium for the
    desktop/workstation
  • 5 10 years out
  • Rethinking conventional microprocessor
    architecture/design
  • Still decode x86 instructions into IA-64
    operations
  • But what if the decoder is efficient enough
  • and clock speeds are high enough?

14
Making Itanium Fast
  • Two Approaches to µprocessor design
  • Wide Low clock speed, incredibly parallel, lots
    of execution units (e.g. Itanium)
  • Narrow Higher clock speed, simpler core, fewer
    execution units (e.g. Pentium 4)

15
Whats wrong with Wide?
  • Wider is only better if you take advantage of it
  • Feeding a very hungry child

16
Whats wrong with Narrow?
  • Which is faster?
  • ADD 22, then, MULT 4 x 4, or
  • ADD 22 and MULT 4 x 4 at the same time
  • The key is parallelism
  • Case in point Intels Hyper-Threading AMDs
    Multicore K8
  • Both AMD and Intel recognize this, but how do we
    make something very parallel and very high
    frequency?

17
The Dream High Freq Wide CPU
  • First ask, why cant wide CPUs run at high
    frequencies?
  • How do we feed a wide CPU? By giving it as many
    instructions in parallel as possible.
  • Problem 1 Data dependencies between
    instructions
  • Can we execute the following simultaneously (in
    parallel)?
  • A B 5
  • C A 6
  • Nope! Instruction 2 is dependent on the outcome
    of Instruction 1.

18
High Freq Wide CPU (continued)
  • How do we get around unavoidable data
    dependencies?
  • Techniques already in place to overcome a good
    deal of data dependencies
  • Solution for very wide processors very large
    scheduling windows

19
Problem with Large Windows?
  • Whats wrong with large scheduling windows?
  • Very slow
  • Limit clock speed potential there goes our
    dream of a 10GHz wide CPU
  • Cant move to a small scheduling window
  • Well hit high clock speeds, but
  • cant fill those execution units
  • Back to square one, is it impossible to build a
    high frequency, wide CPU?

20
Hierarchical Scheduling Windows
  • One potential solution in the research phase
  • Hierarchical Scheduling Windows (HSW)
  • Instead of having one scheduling window, lets
    have two
  • 1 small, fast scheduling window
  • 1 large, slow scheduling window
  • Each window gets its own cluster of execution
    units

21
How HSW Works
  • Small, Fast window gets instructions that arent
    latency tolerant
  • i.e. impatient instructions that must be executed
    immediately
  • Allows for the characteristics of a high
    frequency core to be maintained
  • Large, Slow window gets instructions that are
    latency tolerant
  • Allows for the benefits of a large window to be
    had without hampering clock speed
  • Feeds hungry execution units

22
The end result?
  • HSW is one step in enabling
  • Very wide,
  • Incredibly parallel,
  • High Frequency microprocessors (10 GHz)
  • But thats only part of the story
  • Manufacturing processes
  • Packaging
  • Branch Prediction Algorithms
  • Multicore technology
  • And much more
  • Will this enable Itanium on the desktop?

23
Thank you for your time
  • Check us out on the web at www.anandtech.com
  • For more about HSW read E. Brekelbaum, J. Rupley
    II, C. Wilkerson and B. Black, Hierarchical
    Scheduling Windows
Write a Comment
User Comments (0)
About PowerShow.com