Wanna_bmw:
Can't seem to leave your rose colored glasses off for even a minute. P4 is lousy at current x86 software. One of the huge selling points that Intel used to extract server market share was the wealth of applications already running on x86 systems. This vast wealth of applications, that was optimized for Pentium Classics, IIs and IIIs, runs very poorly on P4, but very well on Athlons and its derivatives. Now Intel has thrown that baby out with the bath water by attempting for the fourth or fifth time to switch architectures to IA-64 from x86. It will not be able to use as a selling point that vast supply of x86 applications and whereas a recompile could help P4, a rewrite is required to make it run well on a IA-64 box (still hearing of compiler troubles with IA-64). That kind of thing usually nets you about 5 to 10% of the market which usually chooses the software and then, if necessary, the hardware. Pioneers can be very fluid. The rest of the market will fight before switching.
This is not AMD's problem. All the wealth works without a change. Testing of that wealth takes about 6 months to a year for compatible hardware. For incompatible hardware, add 1 to 4 years, if ever. A great performance or a great price may shrink the time for the ramp, but testing still takes the same amount of time. X86-64 is compatible. It does not need IA-64's lag time of a year or more.
Given that for untouched x86 applications, P4 has between 33 to 66% of the performance at same clock of P3. Against Palomino, 25% to 50% of the performance at same clock. IA-64 in compatibility mode has 5% to 10% of the performance of P3 at same clock. Have not seen enough native mode IA-64 real world software times to compare.
X86-64 has three big reasons for superior desirability of existing customer base. X86 applications work at full speed completely compatible (just plug in and use). Hammer will have much higher base performance (x86-32). Hammer allows for a 64 bit mode that is piecewise upgradable (start with 64 bit OS, and move applications over one at a time when they are ready). A bonus is that 64 bit performance will be even better than the already great IA-32 bit performance (flat register set, twice the GPRs and SSE(2) regs and no more segment registers (easier for a compiler to optimize)).
IA-64 will be dropped even faster than it was pioneered. After taking a brightly lit four lane interstate off ramp, IA-64 pioneers have faced a road that turned into two lanes, then became twisty, graveled with under construction signs, numerous partial washouts, centered with grass and lastly two mud tracks lit with construction beacons terminating with a bridge with signs of McKinley Bridge and FUD river. The other side is shrouded in dense fog and what you can see of the bridge is strewn with gaping holes, rebar and lumber. Next to the bridge is a interstate on ramp signed x86-64 Hammer. It is well lit and looks just like what you left way back there going twice as fast or more. What do you think most pioneers will do? You guessed it, they will leave the IA-64 way in droves saying "they don't pay me enough for this sh--!"
They have found out about AMD in notebooks and left Intel in droves. You hope it will not happen with IA-64. P4 Xeon may have more chance, but don't bet on it.
Pete |
