Alf Peake <alf### [at] peake42freeservecouk> wrote:
: Working myself up for a rebuild and having always used Intel, I'm now
: convincing myself that perhaps an Athlon is worth a try this time.

  AMD made huge improvements to the so-called P6-family of processors
(P-II, P-III, Celerons based on those and all Athlons and Durons all belong
to this processor technology type, with just minor differences in details)
for its Athlon processor. (Although Athlon is called K7, it really is just
another P6 processor; the change in number is quite misleading.)
  Athlon could be called the top-of-the-line P6 processor. AMD made huge
efforts in eliminating some big problems of previous P6 processors (many of
these problems made the processor slower than necessary, eg. by causing
unnecessary stalls or penalties, one example being the so-called partial
register stall, which AMD succeeded in removing from the Athlon).
  Athlon is practically as fast as a P6 processor can be (if we count the
number of executed instructions per clock).

  The Intel Pentium 4 uses a truely newer technology. However, Intel made some
really strange compromises and perhaps even mistakes (whether intentionally or
unintentionally is a mystery) which makes Pentium 4 a really slow processor
(when counting as executed instructions per clock). Pentium 4 can try to
compete with Athlon only because it can use a 50% higher clock rate and also
because it uses only the highly expensive type of faster memory. But even
then it doesn't really match an Athlon with one third slower clock rate and
slower memory (and half of the price).
  Pentium 4 really wastes huge amounts of clock cycles for things that should
be faster. For example a simple shift operation (which is used in
compiler-generated code a lot) has taken 1 clock cycle since 486 (up to P-III),
but in Pentium 4 it can take something like 8 cycles.
  Even when a compiler can optimize specifically for Pentium 4 and it can gain
a lot of speed, it still doesn't get all the speed it *could* get for that
clock rate. Practically the optimizing compiler has to avoid all those
compromises which make the code slow (eg. shifts).

  The new technology might be promising, but the current implementation is
just a waste of money. Don't waste your money in it.

-- 
#macro M(A,N,D,L)plane{-z,-9pigment{mandel L*9translate N color_map{[0rgb x]
[1rgb 9]}scale<D,D*3D>*1e3}rotate y*A*8}#end M(-3<1.206434.28623>70,7)M(
-1<.7438.1795>1,20)M(1<.77595.13699>30,20)M(3<.75923.07145>80,99)// - Warp -

Post a reply to this message