Search the FAQ Archives

3 - A - B - C - D - E - F - G - H - I - J - K - L - M
N - O - P - Q - R - S - T - U - V - W - X - Y - Z
faqs.org - Internet FAQ Archives

SGI hardware Frequently Asked Questions (FAQ)
Section - -8- How fast is my R4000 or R4400 machine?

( Single Page )
[ Usenet FAQs | Web FAQs | Documents | RFC Index | Business Photos and Profiles ]


Top Document: SGI hardware Frequently Asked Questions (FAQ)
Previous Document: -7- Should I shut off my Iris at night?
Next Document: -9- What is the IP number of each SGI model?
See reader questions & answers on this topic! - Help others by sharing your knowledge

  Eric Williams <williams@agomoda.asd.sgi.com> reveals all:
  It is confusing to spec the clock frequency for the R4000 and R4400
  because they are so flexible. There are four interesting numbers:

    - internal clock
    - external clock
    - secondary cache access cycle
    - SYSINT frequency

  Let's start by specifying the processor internal frequency.  e.g. 150
  MHz.  All other frequencies are specified with respect to this one.
  For programs that get good primary cache hit rates this number will
  determine the performance.

  The clock input to the R4400 (i.e. the crystal you buy) is always
  half the internal frequency.  In this case 75 MHz.  This is generally
  the number used by the chip manufacturers, to specify the speed of
  the part.  However from a system point of view, it is the least
  visible to the user, and therefore IMHO the least interesting.

  The secondary cache read and write access cycles are programmable in
  terms the internal clock frequency (e.g. 150 MHz cycles).  This
  allows you to trade off the cost/speed of secondary cache rams with
  system performance.  When upgrading from 100 MHz to 150 MHz you can
  either keep the same rams and increase the SCache access cycle or
  install faster rams and keep the number of cycles constant.  The
  first option keeps the cost to a minimum while the second maximizes
  performance.

  Finally the interface that talks to the system (SYSINT) can run at a
  programmable fraction (1/2, 1/3, ...) of the internal frequency.  For
  the example 150 MHz processor, this could be 75 MHz, 50 MHz, etc.
  This puts an upper limit on the bandwidth to memory and affects some
  latency parameters.  Typically you would program the system interface
  to run synchronously with the memory controller.

  From what I've heard here about the Indy R4400 upgrade (I'm not
  involved with it) I think you could say the following:

    - the internal clock (primary cache, instruction execution, etc)
          increases from 100 MHz to 150 MHz
    - the clock crystal increases from 50 MHz to 75 MHz
    - the secondary cache access times stays the same in absolute
          terms (but increases in terms of internal clock cycles)
    - the system interface to memory stays at 50 MHz (100 MHz div 2, vs.
          150 MHz div 3)

  BTW, the Indy upgrade example illustrates why IMHO the 75 MHz
  external frequency of the R4400 is not an interesting number to
  quote.  Performance of real programs will be determined by the
  internal 150 MHz clock, the secondary cache timing and the system
  interface/memory speed, not the 75 MHz external clock.

  The Jan/Feb 1996 Pipeline has a table of Indigo, Indy and Indigo^2
  processor types and the versions of IRIX which support them.

User Contributions:

Comment about this article, ask questions, or add new information about this topic:

CAPTCHA




Top Document: SGI hardware Frequently Asked Questions (FAQ)
Previous Document: -7- Should I shut off my Iris at night?
Next Document: -9- What is the IP number of each SGI model?

Single Page

[ Usenet FAQs | Web FAQs | Documents | RFC Index ]

Send corrections/additions to the FAQ Maintainer:
sgi-faq@viz.tamu.edu (The SGI FAQ group)





Last Update March 27 2014 @ 02:12 PM