> Funny enough, when I wrote and
> said that all Intel CPUs that have floating point support in hardware
> (i.e. 8087, 80287, 80387, 486, Pentiums) all use 80-bits internally, and
> 64 bits to store the number in memory, that letter was never published.
Not quite right. The newer processors have the ability to do both IEEE 80
bit (ala 8087) and newer instructions (e.g. SIMD extention 2). The new 64
bit processors have new floating point capabilities, new registers,
including SIMD 2, new instructions, the ability to store a float in a single
write (as opposed to two 32 bit operations), in fact, most of them can do
128 bit writes in a single cpu/cache or cache/memory operation. It is these
additional capabilities that are targeted towards multi media processing
(and games), and they make a substantial difference when used correctly on
the correct workload. They also have vector like parallel processing
capabilities AND 128 bit media an scientific instruction sets. When applied
to multimedia, video and numerical workloads, these can be dramatically
faster than x87 style single stream floating point math. I use these
capabilities at work, and they make a big difference.
Incidentally, the assertion that for '99.9% of computer users, a 64-bit
computer is a complete and utter waste of time', is not completely accurate,
though a perfectly valid opinion. . There is an old engineering expression,
"The number of people swimming across the river today is not an accurate
estimate of the load on the bridge you want to build tomorrow". Part of the
marketing research that semiconductor manufacturers do before investing
billions in a new processor line considers what future processing demands
would require. It seems very likely that the vast majority of the market
would be interested in the multimedia capabilities that these kinds of
processor enhancements make possible. Areas like games (huge number of
users), video content searching, voice recognition, cryptographic
functions, photographic recognition, 3d graphics, other multimedia
capabilities etc, are pushing the limits of what exists in the current 32bit
offerings. There have been two approaches; one is to offload the processing
to a special purpose processor (e.g. DSP etc), the other has been do perform
the processing on the main CPU. Again and again, when we've looked at the
economies of scale, doing these things on the main processor is frequently a
big win for all parties. But we're starting to hit the wall on the 32 bit
processor families. So it really is more than just 'making the previous
processors obsolete'.
The other interesting dynamic is that as the 64 bit wave arrives in the mass
market, it will cannabalize 32 bit processor business, and 32 bit processor
prices will drop precipitously until end of life. This will make it possible
to use the (very inexpensive) 32 bit processors in areas where their cost
was prohibitive.
Also, this may spell the end for other processor families, eg Itanium, and
Sparc 64 (Sun is already heavily into AMD Opteron) as the volume on the AMD
Opteron/Intel EM64T processors ramps up, competing with these processors
will be an even tougher business than it is today.
So, if you assume that most people don't need a 64bit memory address space
today, that's probably true (although the local high school kids need more
than 4 GB because they're doing audio recording/sampling/mixing on PCs). But
if you think at what becomes possible with the newer processors, it no
longer seems such a complete waste of time.
73's
Ed
----- Original Message -----
From: "David Kirkby" <david.kirkby@onetel.net>
To: "John T. M. Lyles" <jtml@lanl.gov>
Cc: <amps@contesting.com>
Sent: Friday, August 05, 2005 3:21 PM
Subject: Re: [Amps] audio black magic TSPA
> John T. M.>> http://www.g8wrb.org/useful-stuff/audiophools.pdf
> Lyles wrote:
>
> > Following is an excerpt from a 1998 Stereophile
> > magazine review of a particular Kinergetics power
> > amplifier, to explain what I am talking about:
> >
> <snip>
> But that is most unlikely to happen again.
>
> The very valid point made by the author of
>
> http://www.g8wrb.org/useful-stuff/audiophools.pdf
>
> is that the publishers of audiophool magazines have a conflict of
> interest. It is not in their interest to tell you what a waste of money
> the speaker cables are when the companies selling to the audiophool
> market provide the publishers advertising revenue.
>
> It's like all the computer magazines tell you how you need 64-bit
> computers. None seem to point out that for 99.9% of uses, a 64-bit
> computer is a complete and utter waste of time. There was a 5 or so page
> article in Personal Computer World about a year ago, telling us that
> 64-bit computers do floating point calculations needed in games faster
> and more accurate than 32-bit computers. Funny enough, when I wrote and
> said that all Intel CPUs that have floating point support in hardware
> (i.e. 8087, 80287, 80387, 486, Pentiums) all use 80-bits internally, and
> 64 bits to store the number in memory, that letter was never published.
>
> It is clearly in the interest of computer magazines to push 64-bit
> computers, so their advertisers will place adds in magazines telling you
> how good they are. No computer company is going to place adds in
> magazines that are telling you that buying a 64-bit machine is a waste
> of time for the vast majority of people.
>
> PS
> I am using a 64-bit computer, that's around 7 years old.
>
> --
> David Kirkby,
>
> G8WRB
>
> Please check out http://www.g8wrb.org/
> of if you live in Essex http://www.southminster-branch-line.org.uk/
>
>
> _______________________________________________
> Amps mailing list
> Amps@contesting.com
> http://lists.contesting.com/mailman/listinfo/amps
_______________________________________________
Amps mailing list
Amps@contesting.com
http://lists.contesting.com/mailman/listinfo/amps
|