<< Previous Message Main Index Next Message >>
<< Previous Message in Thread This Month Next Message in Thread >>
Date   : Wed, 29 Jul 2009 21:29:34 +0200
From   : rick@... (Rick Murray)
Subject: Units of memory, Was: Master Ethernet upgrade

Phill Harvey-Smith wrote:

> it makes much more sense to talk in something that is a power 
 > of 2 e.g. 1K=1024, 1M=1024x1024 etc, this is what I use :)

> However there are those that would argue that in SI, K=kilo=1000, and 
> strictly speaking they would be correct.

I still believe it all came about from a marketing con. For all eternity 
(i.e. before zero Unixtime) we used kilobyte and we all knew it meant 
1024 because a power of two is logical. Powers of ten only make sense to 
a decimal computer and I'm not aware of such a thing (though no doubt 
somebody has tried...).

Nowadays? My 250Gb harddisc is something like 232Gb if measured 
'correctly'. I say correctly in quotes because the drive understands 512 
byte sectoring, not 500. It understands IDE addressing, which by their 
very nature is going to be binary powers, not tens. It will return bytes 
or words, not 10 bits and 20 bits. Yet suddenly when we talk of its size 
  we're all base ten. Just so the sales bod can say "whoo! look, it is 
250Gb" while handing me a unit that is more shy of 250 actual Gb than 
the size of the 'harddiscs' in Azumi!

It's only sad that in *other* circles, where units of 10s/1000s make 
more sense, the same prefixes are officially designated, thus being a 
convenient argument to the flamin' illogical. Show me a decimal harddisc 
that has 1000 byte sectors where each byte is 10 bits. Exactly...


> I once had an argument with a friend at school, where I argued that you 
> needent used 1 and 0 to represent binary, you could use any pair of 
> symbols as long as you defined what they meant, X and Y for example so 
> say $5A would be XYXYYXYX :) he wouldn't have it and isisted that it had 
> to be 0 and 1.

He is *SO* wrong it is actually funny! Zero and one is simply a useful 
way we have of writing it, based purely upon the first two digits of our 
counting system. However zero and one means diddly-squat to a processor 
or memory unit. It's presence of charge or absence of charge. Hook an 
LED array to the data bus, it will be easier to then think of it as 
on-off-on-on-off-on-on-off instead of 10110110 or even 
five-zero-five-five-zero-five-five-zero (assuming TTL! <grin>).

Back in the old days, we all used sticks and stones, which I guess can 
be easily translated to zeros and ones - round stones and stick-like 
sticks... :-)


Best wishes,

Rick.


>> Are you sure? Why would you use mmA when there's a perfectly usable
>> uA?
> I believe from what I have read that was common in the USA at one point.

Perhaps mmA (which I have seen a while ago, didn't know what it meant!) 
is for anal people that can't just write "uA" in place of the correct "?A".


> Humm tell that to battery manfacturers that insist on quoting 
> rechargables as 2500mAh, instead of 2.5Ah,

2-point-5 is mundane. TWO THOUSAND FIVE HUNDRED is massive!


> To drag things a little back on topic compare the BBC and the Spectrum, 
> going on clock speed the Spectrum should be better as it has a higher 
> clock, approx 3.5MHz as opposed to 2MHz, but anyone has actually used 
> both machines will tell you which is faster for most things...

The 6502 will outperform most of the other 8 bit processors if you 
consider equal clock speeds. The only reason the Z80 is used a lot, I 
feel, is because it come in a much greater range of speeds. I have a 
little printer buffer gizmo that runs a Z80 at around 8MHz (or was it 
six?). I bet if we had it running a 6502 at that sort of speed, it would 
outperform. Certainly the 6502 flies when it comes to interrupt latency. 
It takes longer on the Z80, and - god - we won't even discuss how long 
an 8086 takes (while the ability to address a bigger gob of memory is 
possibly what made it the CPU of choice in proto-PCs, what's with the 
weird clocking cycle?).


Once upon a time it was possible to quote PC clock speeds for a show of 
superiority (and less so with the RiscPC, though you really only had a 
choice of ARM610, ARM710, various StrongARM, and Kinetic). Nowadays it 
seems CPUs have maxed out around 2.8-3GHz and instead all sorts of 
trickery comes in to play (dual core, etc [*]), and the model numbers 
bear no relationship to the speed. Not only that, if you look at 
comparative specs, your CPU is perhaps less important than the FSB. For 
example, "Aiko" runs a 450MHz P3 on a FSB running around 100MHz. 
"Angelique" (the old laptop) runs a 466MHz Celeron on a FSB at 66MHz. In 
tests (namely the fps possible coverting MPEG1 to XviD), Aiko is about 
double the speed of Angelique despite the slower processor.

For those who know of neither PC design nor WTF I call my computers, a 
perfect Acorn analogy is "put a Kinetic in your RiscPC and tell me how 
much faster your SCSI card runs". It won't. In fact the Kinetic stuffs a 
bunch of memory on the CPU card in order to work around the bottleneck 
that is the RiscPC's 'FSB'. What is it, 12MHz or something? Running a 
200+MHz processor? :-)
Mmmm... if the original ARM610 could do something like 30MHz (I'll need 
to find mine to be sure), surely the RiscPC's FSB was 'too slow' right 
from day one? The A5000 almost manages that bus speed. Didn't the A540 
run its memory bus at 12MMHz or so?

Life was easier in the Beeb era. You had:
   Slow   - Electron (CPU/video memory access dohickeys)
   Medium - Beeb
   Fast   - Master 128, thanks to CMOS processor
   Insane - Beeb with a 65C02 co-processor
:-)


Best wishes,

Rick.


* - Azumi's "Atom" processor is "hyperthreading" as opposed, I presume 
to dual core. Does this mean, then, that there is only one actual 
processor core while the thing has lots of trickery and magic to make it 
look like two? If I Ctrl-Alt-Del I can pick a process and set its 
affinity to either CPU0 or CPU1 or both (I like how the window goes up 
to CPU31!). Running ProcessExplorer, it can show separate activity 
charts for each 'processor', which can vary from each other - yet all I 
am aware of on this CPU is "hyperthreading", not "dual core"...?

-- 
Rick Murray, irregular internet access at local library.
BBC B: DNFS, 2 x 5.25" floppies, EPROM prog, Acorn TTX
E01S FileStore, A3000/A5000/RiscPC/various PCs/blahblah...
<< Previous Message Main Index Next Message >>
<< Previous Message in Thread This Month Next Message in Thread >>