benhoskings at hotmail.com
Sun Oct 20 03:48:21 PDT 2002
Ian Molton wrote:
> On Fri, 18 Oct 2002 18:59:35 +0000 (UTC)
> JDrabb at darden.com ("James Drabb") wrote:
>>>Actually, the A7000+ is an OK performer - its no games machine, but
>>>it'll do better desktop publishing than a PC with 10 times its spec.
>>Still, how fast can a 300 MHz RISC be?
> Quite nippy actually. Bear in mind a lot of RISC OS is written in
> assembler by people who KNOW how to write assembler.
>>It can still only do at most 300 million ops per second.
> Wrong. you can actually do up to about 2.5 times better than that as the
> later ARMs are superscalar (I think thats the term used for a CPU
> capable of executing >1 insn at once, I always mix it up with some other
> For example, on some ARMs, a 32 bit MUL takes 3 cycles and you can
> execute any non-MULL non-conflicing instructions in the following 2
> cycles (and its actually possible to do so on ARM as its register rich
> (15 general purpose, and the program counter available to usermode
>> Versus a cheap celeron/duron
>>running at 1.x GHz or a P4 at 2.x GHz.
> True, a 1-2GHz cel/dur will blow it away in raw grunt, however, compared
> to a 300MHz part, the StrongARM does very well indeed.
> the SA pipeline is only 5 entries long, and branches take a MAXIMUM of 3
> cycles (if in the insn. cache, of course). loads / stores from cache
> take 1 tick. it has powerful multiregister load/store operations too.
>> Was that a typo or do
>>they really have 64 MB of L2 cache or did you mean main memory?
> Both, sorta. the 'kinetic' ones (with the 300MHz CPU) have 64MB of RAM
> on the card, connected to the CPU at 66MHz. the main memory bus
> (remember the original machine is over 8 years old) is only 16MHz. the
> on-cpu-card RAM could be considered L2 cache, is a loose sense. hence I
> wrote it in invert commas.
> The thing is, though, that RISC OS programmers have simply tended to be
> very good - for example, my **8** MHz ARM2 based system (ARM2 has *no*
> cache) with shared video and system RAM, now 15 years old, is capable of
> rendering its *ENTIRE* GUI in antialiased vector fonts, at any scale,
> resolution, or orientation. AND it can do it at a useable speed. Its
> video chip has NO acceleration functions, to boot.
> compare to KDE / gnome2 on current hardware, just barely managing the
> same task...
> Did I mention, the old Acorn was doing *subpixel* antialiasing?
> Impressive machines, despite lacking raw grunt. The only lacking thing
> is an FPU, but ARM hackers have done very well without it - eg. an MP3
> player running in <5% CPU time on a pure integer machine without
> halfword stores (which strongarm supports, but the RISCPC cant). FPUs
> give one benefit - superscalarity. its proven that they dont actually
> improve performance over well written integer / fixed point code.
Jeez that's cool. I'm doing a subject at Uni this semester on Computer
Architecture, so I now know about the program counter, instruction cache
and all that you mentioned then. Man, I'd love to have one of those
machines to play with - it sounds danm interesting. And as you say,
although it might not have the raw grunt of the current x86 flagships,
it sounds like a REALLY nicely designed arcihtecture, and some REALLY
damn good asm programming. :D
Anything that can render a whole GUI of AA fonts entirely in software,
with no cache, at 8MHz (!!!) has my seal of approval. :)
I want one!
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-chat' in the subject header of the message
More information about the lfs-chat