From: Mika Liljeberg (liljeber@hydra.Helsinki.FI)
Date: 04/30/93


From: liljeber@hydra.Helsinki.FI (Mika Liljeberg)
Subject: Re: Intel, the Pentium and Linux
Date: 30 Apr 1993 15:38:23 GMT

In article <willmore.736143868@help.cc.iastate.edu> willmore@iastate.edu (David Willmore) wrote:

> liljeber@hydra.Helsinki.FI (Mika Liljeberg) writes:

>>This is not true. All modern processors have an onboard instruction
>>cache. A RISC processor doesn't waste any more memory bandwidth than a
>>CISC processor. If we assume that a processor can perform one load or
>>one store / clock cycle (not a rule perhaps, but certainly the usual
>>case), the RISC processor with its shorter clock cycle has a definate
>>advantage here.
>
> You have to fetch them sometime. It's going to cost more than a CISC.
> Not much, sure, but it will cost more.

Yes. With loops the difference is negligible, though.

> Also, A CISC has more exposed
> parralelism so that if you have a 2 access/cycle data cache, the CISC
> can take advantage of it more easily. A RISC processor would have to go
> superscalar or superpipelined to be able to do that.

Hmm. This sounds logical on the outset, but such simplicity is one of
the _goals_ of the RISC architecture. This means that a RISC chip
doesn't need such a sophisticated cache in order to realize its full
power. On the other hand, a superscalar RISC chip can use a 2
access/cycle cache to the fullest, while it is largely wasted on a
CISC chip. Not every instruction uses complex addressing modes, after
all. A superscalar CISC, now, would need an even more sophisticated
cache in order to exploit the "exposed parallelism" of memory
accesses.

The point I'm trying to make is that RISC chips simply work better
with caches, because memory accesses are more evenly spaced. Also, it
is much easier to design superscalar RISC chips than superscalar CISC
chips. It's not that the problems with CISC are insurmountable. It's
just that they are costlier to overcome. Certainly Intel has shown
that CISC is still a viable alternative.

But enough on this vein. The relative merits of RISC and CISC have
bene talked to death many times over, and this isn't really the proper
newgroup for it.

> Later,
> David
> --
> ---------------------------------------------------------------------------
> willmore@iastate.edu | "Death before dishonor" | "Better dead than greek" |
> David Willmore | "Ever noticed how much they look like orchids? Lovely!" |
> ---------------------------------------------------------------------------

        Mika