Replies to post #101614 on Intel Corp (INTC)

[chipguy]

The core power cost of x86 over RISC depends on issue width,
pipeline depth, and in-order vs OOOE. The front end decoder
complexity goes up with the square of issue width for x86 vs
roughly linear for RISC. OOOE is also more complex for x86
than RISC. For something like C2D/Nehalem/SB x86 probably
adds 25-30% in power over an equivalent RISC. For something
like Atom it probably adds 15-20% in power over an equivalent
RISC.

Now keep in mind for a 1W phone SoC the CPU core accounts
for 300-400 mW. So x86 adds 45 - 80 mW in full active power
when EVERYTHING ELSE IS EQUAL. Please note the qualifier.
It is generally not true. In fact it is so not true that despite all
its baggage x86 accounts for over 90 cents out of every dollar
spent in the world on microprocessors of all size, description,
and use.

[Tenchu]

David, with regard to ARM vs. x86, the same arguments were made back when PowerPC was going to clean Pentium's clock.

It's RISC vs. CISC all over again. That battle ended up with RISC processors becoming more CISC-like and CISC processors becoming more RISC-like. But the ultimate victor was CISC.

Same thing will happen with ARM vs. x86. We already see Intel (and maybe AMD) push x86 into smartphones and tablets. No reason why it can't. Meanwhile ARM and its advocates are trying to push upward into laptops and even into servers and workstations.

I repeat: There really is no reason why x86 can't compete in smartphones and tablets. All x86 instructions get translated internally into uops anyway. That translation takes maybe 5-10% of the CPU die and power. Nothing that can't be made up for by Intel's manufacturing edge and experience in designing for lower power.

Like Chipguy said, all else being equal, ARM should theoretically be better than x86, but things are never equal.

Tenchu