home *** CD-ROM | disk | FTP | other *** search
- Path: soap.news.pipex.net!pipex!usenet
- From: m.hendry@dial.pipex.com (Mathew Hendry)
- Newsgroups: comp.sys.amiga.programmer
- Subject: Re: Purpose of RISCiness (was: FWD: Fate of 68080)
- Date: Mon, 29 Jan 96 10:26:37
- Organization: Private node.
- Distribution: world
- Message-ID: <19960129.42C4A8.9A02@am174.du.pipex.com>
- References: <4e7rhi$4fo@maureen.teleport.com> <4ealme$8fi@hades.datashopper.dk> <9eU-x*Oxf@yaps.rhein.de>
- NNTP-Posting-Host: am174.du.pipex.com
- X-Newsreader: TIN [AMIGA 1.3 950726BETA PL0]
-
- Arno Eigenwillig (arno@yaps.rhein.de) wrote:
- : In article <4ealme$8fi@hades.datashopper.dk>, Michael Berg writes:
- :
- : > (For the slower readers -- RISC means Reduced Instruction Set, which in
- : > turn means the processor has to execute many more instructions than a CISC
- : > architecture to achieve the same thing. I don't recall the typical
- : > RISC/CISC factor, but it's certainly more than 2).
- :
- : Describing it this way neglects the advantages of a RISC architecture:
- : Few and uniform opcodes make the CPU simpler, which allows for smaller
- : chips, thus higher clock rates (think: heat!) and possibly hard-wired
- : CPUs instead of microcode machines. (A microcode processor like any
- : 680x0 has a low-level program in it which is directly understood by
- : its hardware and interpretes the machine code.)
-
- Except that for most modern RISC chips the complexity (in terms of number of
- transistors, and hence power consumption) is not much different to that of
- current CISC chips. The main difference is the _emphasis_ of that complexity.
- Instead of expending design effort (and COST), transistors, and power on every
- instruction / addressing mode imaginable, they devote it to optimising the
- execution of simpler instructions, through the addition of superscalar
- pipelining, larger caches, sophisticated branch prediction logic, optimised
- (in terms of speed) floating point and integer maths instructions, etc. etc.
-
- The current versions of the DEC Alpha have much larger numbers of transistors
- than any CISC CPU (except for the P6, which comes pretty close). Much of this
- space is devoted to huge caches and sophisticated pipelining control and
- sheduling. The fact that the Alpha is still the fastest production CPU on
- Earth is an indication that this is a winning strategy.
-
- : Furthermore, simple machine instructions make it easier to construct
- : parallelized CPUs, because keeping track of dependancies between
- : instructions (which is the main problem with such designs, AFAIK) gets
- : easier.
-
- Yes, and it also leaves you with more die space with which to implement
- this.
-
- People also argue that RISC is a bad strategy because it increases RAM
- requirements. It does, but the fact is that, in real terms, RAM is not that
- expensive these days. A memory-related factor which IS becoming more and more
- significant is the ever increasing gap in speed between memory technology and
- that of CPUs. Without very expensive static RAM chips (which are subject to
- delays anyway if they're off-chip), the RAM just can't keep up. The load /
- store architecture of RISC CPUs, coupled with a large number of on-chip
- registers and advanced caching, goes a long way towards helping the technology
- to escape from this bottleneck.
-
- -- Mat.
-