home *** CD-ROM | disk | FTP | other *** search
- Path: comma.rhein.de!serpens!not-for-mail
- From: mlelstv@serpens.rhein.de (Michael van Elst)
- Newsgroups: comp.sys.amiga.advocacy,comp.sys.amiga.misc
- Subject: Re: Walker vs. the $999 7200/75 w/4xCD!
- Date: 2 Apr 1996 01:24:53 +0200
- Organization: dis-
- Message-ID: <4jpok5$bu4@serpens.rhein.de>
- References: <eraugust-2803960910170001@sbdsk0838.sbrc.hac.com> <1350.6662T1401T2752@mbox.vol.it> <4jpcr3$6ni@valour.pem.cam.ac.uk>
- NNTP-Posting-Host: serpens.rhein.de
-
- cbrown@armltd.co.uk (Chris Brown) writes:
-
- >Your assertion is silly. One of the main reasons why RISC processors
- >came about in the first place was that memory was too slow for CISC
- >processors, with thier instructions that operated directly on
- >memory.
-
- Ugh, ugh, ugh. RISC CPUs usually need faster memory (and more).
-
- >RISC processors have as their only memory operations loads and
- >stores, with lots of registers to keep their data in.
-
- Sure.
-
- >This serialises
- >memory accesses more which means burst mode can be very well utilised,
- >and is kinder on caches.
-
- No. It just makes it possible to decouple operations from memory accesses.
- So if the load or store _still_ takes the same time you can do something
- sensible in the meantime. This is not possible if you operate directly
- on memory.
-
- But RISC needs more and faster memory and this became feasible with the
- development of cheap cache technologies.
-
- >Compare with, say the P6, a CISC chip, which
- >needs an enormous and fast level 2 cache to perform well.
-
- That's exactly what RISC CPUs need. Most high end risc machines come
- with several _megabytes_ of cache. Most Pentiums use 256k or 512k. The
- P6 isn't better in this regard but (like the latest Alpha) has another
- hierarchy level. The external cache there becomes L3 because the chips
- have integrated a relatively small L2 cache themselves.
-
- >>Untill someone makes 5ns main ram (!) and RISC's can go out again, untill
- >>we get BiCMOS technology for CPU's and at 700Mhz and 20 millions of transistors
- >>CISC wins again in any case.
-
- >Huh? You are aware that one of the purest RISC designs around today is
- >the DEC Alpha, right? It's also the fastest commercially available
- >microprocessor by a *long* way.
-
- It is. But all the RISC technologies depend on memory speed. If memory were
- ultimatively slow then CISC were the way to go, the more you can do with the
- single memory fetch, the better. Currently the memory technology can barely
- keep up with CPU technology. Machines already need huge and expensive caches
- (most Alphas have 2 or 4MByte cache, SGIs about the same) and we _may_ eventually
- see the return of more CISCier CPUs. This doesn't mean that we see 68k or VAX
- style machine codes again. These weren't designed to utilize memory bandwidth
- but to make assembly programming easier and CPU designers won't forget their
- RISC lessons.
-
- But that's speculation. There are a couple of memory technologies that can push
- RISC designs even more (and even CISC CPUs will benefit from it, although not that
- much).
-
- Regards,
- --
- Michael van Elst
-
- Internet: mlelstv@serpens.rhein.de
- "A potential Snark may lurk in every tree."
-