Re: Book: Inside the Machine

Neil Rieck wrote:
Because of these drawbacks to deep pipelining, many critics of the
Pentium 4's microarchitecture, dubbed NetBurst by Intel, have
suggested that its staggeringly long pipeline was a gimmick - a poor
design choice made for reasons of marketing and not performance and
scalability.
[...snip...]
It's my understanding that this fact was widely known within Intel,
even though it was not, and probably newer will be, publicly
acknowledged.

My understanding differs. Analysis (I don't still have the reference handy, sorry) indicates the pipeline has to get much longer than P4's before pipeline length per se starts limiting performance. It may well be that the decision to make a speed demon rather than a brainiac came from marketing, but the engineers didn't say (even to themselves) "oh well, guess we've just got to fool the punters with a crap CPU", they said "fair enough, we'll make a speed demon that'll have such a high clock speed that it really will outperform the brainiac competition".

What nobody knew at that point was that the race for higher clock speed was about to hit the heat dissipation wall. P4 was designed to climb past 5 GHz and towards the 10 mark; instead it couldn't even reach 4 GHz without a special cooling system or melting the chip.

As confirmation that Intel didn't realize this, look at the Itanium: it has (architectural!) features that only make sense if the designers didn't know performance _per watt_ was about to become a critical figure of merit.

--
"Always look on the bright side of life."
To reply by email, replace no.spam with my last name.
.

Relevant Pages

  • Re: Book: Inside the Machine
    ... It's my understanding that this fact was widely known within Intel, ... Analysis indicates the pipeline has to get much longer than P4's before pipeline length per se starts limiting performance. ... whistles got thrown into the pot (my impression is due to design by committee), Itanic's basic premise is all about simplicity and efficiency. ... The only way the 'heat dissipation wall' that Netburst hit should have been a surprise to Intel's design engineers would have if their process-engineering associates seriously misled them about the anticipated process gains: the more pipeline stages you have the more transistors are involved, and the faster you switch those transistors the more heat each one of them generates, so the power requirements of Netburst at any given frequency should have been fairly easy to anticipate for any well-understood process. ...
    (comp.os.vms)
  • Re: Book: Inside the Machine
    ... instruction completion rates if the pipeline has to be flushed and ... It's my understanding that this fact was widely known within Intel, ... We now know that during the Pentium 4's design, ... design team was under pressure from the marketing folks to turn out a ...
    (comp.os.vms)
  • Re: Segmentation in real mode
    ... Why do segment registers hold segments' paragraph numbers rather ... As to why Intel used _this particular method_ of doing it rather than some ... architectures, other than the Apple Mac...indeed, the last chip they made - ... previous design but not made "100% backwards compatible" with that design, ...
    (alt.lang.asm)
  • Re: IBM Clown "Easter Egg"
    ... that design was either not oscillating at all or taking off at the 2nd ... One of the major frustrations was the single source of the Intel ... licensed NEC, ... The IFM firmware code didn't require ...
    (comp.os.cpm)
  • Re: Another x86-64 price war is coming
    ... AMD is about to release a true quad core chip. ... Depending on what part of the market you are interested in, either the system price is key, or how much can you feed through the design and where the bottlenecklurk can be key. ... With the various Intel 64 designs I've looked at, the traditional technical bottleneck appears to involve the FSB out through the northbridge; bandwidth through that has not matched that of AMD64 designs, nor that of the EV7 design. ...
    (comp.os.vms)