Re: New record for spec fp peak on x86

"Greg Lindahl" <lindahl@xxxxxxx> wrote in message
news:44c2f4e9$1@xxxxxxxxxxxxxxxx
In article <4hAwg.10097$YO6.5081@xxxxxxxxxxxxxxxxxxxx>,
Alexi Tekhasski <a.pred...enski@xxxxxxxxxxxxx> wrote:

Incidentally, I don't recall any reasoning on your part at all at
the time,

It was that only an idiot would look solely at clock speed when it was
known that the core was going to be substantially different than just
a clocked-up version of the same old core.

I must miss your whole reasoning. Could you provide
a link where you have formulated this fascinating thought
two years ago, especially if no "new core" was under
any discussion at the time?

Therefore, your "a lot other than memory subsystem improvements"
does not agree with the experimental facts presented above.

I missed the "facts", I saw an analysis which was just a bit more
sophisticated than showing 1 + 1 = 1 by multiplying by zero in the
"proof".

Published SPEC performance scores are the facts, and
my little obvious analysis on the top of these numbers
does not take any assumptions. Therefore, the elementary
transformation of published numbers holds up the same
strength as facts. You seem to have difficulty either in
comprehending the presented elementary mathematics,
or with interpretations I provided.

Hint: you're combining a lot of non-linear things and
claiming it's obvious that it should be linear, and then that you're
vindicated because it ended up being linear.

Interesting, I'd like to hear more about "non-linear
things" in coupled sequential FSMs as a function
of main clock, and how a sun of many nonlinear
things could end up being linear.

Note, I do not combine anything, computer combines
all that stuff, linear or nonlinear or whatever, and ends
up just as I presented. And I am not claiming that
it is obvious because it is apparently not, for some
at least. However, this is what every benchmark is
doing for the last decade, on all evolving platforms,
as all published data indicate, and in contrast with your
opinion.

I think the point of your confusion is that you accustomed
to look at a workload from a microscopic point of view,
individual loop analysis, etc. As I stated many times,
my anaysis is at a macroscopic level. It does not show
details of microexecution, but rather shows a big,
user-level picture, and points to areas of trouble that
might be non-obvious otherwise.
I am sure you are familiar with an expression about
a forest missed behind the trees...

That's like claiming that
a few years of great stock picks means that you're a good stock
picker, instead of just another gambler.

Do you allude here that your method of fiddling with
compiler options is no different from gambling on
stocks? :-)

I'm not upset, and I don't "win" or "lose" when arguing with loons;
I always lose when wasting my time like this.

It is certainly a waste of everyone time to argue against the
fundamental law of performance that is a corollary from well
established Hennessy and Patterson performance equations.

- Alexi

.

Relevant Pages

  • Re: New record for spec fp peak on x86
    ... a clocked-up version of the same old core. ... claiming it's obvious that it should be linear, ... vindicated because it ended up being linear. ... a few years of great stock picks means that you're a good stock ...
    (comp.arch)
  • Bill Reids post
    ... related) into a neural network, then asking the network to predict each ... Not that neural networks can predict the stock ... This is very different from linear algebra, ... green curve it produces is matching the blue curve, this means, it HAS ...
    (misc.invest.stocks)
  • Re: Proof that neural nets work
    ... related) into a neural network, then asking the network to predict each ... Not that neural networks can predict the stock ... This is very different from linear algebra, ... green curve it produces is matching the blue curve, this means, it HAS ...
    (misc.invest.stocks)
  • Re: Processor upgrade for Core Solo Mini
    ... Merom, Yonah, core solo/duo/core 2 duo... ... Before anyone asks it only has stock 2 x 256MB memory, ... Just do the memory upgrade. ...
    (uk.comp.sys.mac)
  • Re: Game 360 Update
    ... > have 18 premiums and 12 core. ... So your store's stock allocation does seem ... when he actually gets stock today or tomorrow. ...
    (uk.games.video.xbox)
Loading