Replies to post #98589 on Intel Corp (INTC)

[chipguy]

Single core ARM performance begins to improve dramatically,

Intel have been building OOOE superscalar cores for 15 years
while ARM is just starting kindergarten in this field. Think of
all the smart companies trying to build a faster core that
Intel buried over the past couple of decades. So easy to go
off into the weeds of complexity. Go ask Sun. :-)

or 2, the # of cores per processor begins to grow very rapidly allowing total CPU performance to approach that of Intel.

Sorry, I call horseshit. Dozens of start-ups have come and gone
trying to build great chips with dozens or hundreds of wimpy
little cores. It is the semi industry version of the dream of
a car that runs on water.

The problem is more cores, be fast or slow, the harder it is
to get them to talk together without chaos. Intel is decades
ahead of any random ARM licensee OEM trying to get a
whole bunch of cores working together well on one chip.

[mmoy]

> ARM concerns me only in 1 of the following 2 cases... Single core
> ARM performance begins to improve dramatically, or 2, the # of
> cores per processor begins to grow very rapidly allowing total CPU
> performance to approach that of Intel.

My son was supposed to work on a 1,000 core system last year for his research internship. The company that made the system was out of business by the time he started. Why do you think a bunch of little cores hooked up together can do the same workload as a powerful processor?

It can in certain workloads that are amenable to parallel processing. Video transcoding, database servers, web surfing are good candidates. A word processing program is a terrible candidate. Games are good candidates with a given number of cores but you have to design for a particular number of cores.

BTW, Intel has an absolutely top-notch compiler with a parallel framework and auto-parallelizing code generation but even there, there are limits to what compilers can do.