I can't believe a Ph.D can be that blunt.
"You still haven't "learned", Ali ... Even your "proud" ModelHurtz "smoke and mirrors"
can't hide the "process thingy". AMD is behind in process technology again ... and again. <ggg>"
Let me try again:
A CPU performance MHz (if measured in MegaHerz) is
a product of "Architecture Thingy", A, and a
"Process Thingy", P.
MHz = A * P;
The A can be measured in the number of pipeline stages,
N, with a coefficient that reflects the balance quality
of the design implementation: A = cN.
The performance of P can be roughly measured by frequency
of a CMOS gate with some typical load. The P must double
in value with every transition to next generation, if
scaling still works as it should (which is not anymore).
You cannot judge each of the two components separately
by looking at MHz alone.
Now, let me propose a historical analysis of the above formula,
at the level that should be comprehensible even for a Ph.D.
For the Pentium MMX, N was about 5, and K6 had N=6,
so the resulting MHz were about the same at the same
0.35um process - 233-266MHz.
Then Intel jumped into PPro/Klamath core with N=12,
which allowed them to get a product with twice MHz
(533-600Mhz) as compared to K6, solely capitalizing
on the "architecture thingy".
To counter the move, AMD had to grasp the same process generation (0.25) thingy to
the last straw, and managed the K6 up to 500MHz, which
was an extraordinary achievement for the "process thingy".
To get higher, AMD introduced the Athlon with N=10,
which made it to 850 in 0.25, then AMD was first to the
1GHz barrier on 0.18, and now up to 1700MHz. Note,
that for the fixed N, the MHz on Athlon design are
up exactly by a factor of 2, as it should be for a
perfectly-scalable transistor architecture ;-)
At the same time Intel struggled with P-III at N=12,
and could not make it above 1GHz in 0.18, no matter
how tough you say they are in the "process thingy".
I guess their "c" were somewhat sucky inside their
"architecture thingy".
Now Intel has arrived with P4 and N=20, taking an
instant "architecture thingy" advantage, making it
up to 2GHz on 0.18 thingy. Despite of this, the
lack of actual performance of P4 (as compared to
Athlon) forced Intel to accelerate the transition to
0.13um thingy, where they are currently grasping at
straw at 2.4-2.5GHz due to unavoidable leakage problems.
Now, what's next? I think AMD should go for N=20, and
get almost instant advantage in MHz up to 3400 (given
1700 at N=10). Then, using the Intel's own turf of
twisted benchmarking, beat them to dust. That would
be fun to watch. In this case they even do not need to
change anything in their current "process thingy".
Did you grasp now the general idea of mutual relationship
between the two "thingies"?
- Ali |
