[extropy-chat] more moore

[Eugen Leitl]

On Wed, Sep 01, 2004 at 11:07:27AM +0100, BillK wrote:

> I think you need to define desktop computer a bit more clearly here.

Easy: the usual machine you pick up at the retailer for 1-2 k$. It's the same
hardware that's being used for supercomputer clusters, plus magic
interconnect (which doubles the price of each node).
 
> Yes, developer workstations will always want more processing power,

No, developer do fine on office-grade machines, actually. If software build
is the bottleneck, you can always farm that out on a cluster. Few other tasks
are such trivially parallelizable as a software build.

> but they are a small part of the market.

Gaming drives the high end part of the market. Absolute performance, not
crunch/Wattage. Embedded DSP has giant volume, but trails
performance/processes. Look at recommended hardware requirements for current
top of the line codec playback.

> Television and DVD players are a much better medium for watching video
> on. We may have stopped watching 100 channels of rubbish on TV, but

The display part is irrelevant, it's the decoding of the video stream it's
where it's at. Codecs are in rapid flux, and hardware implementations are too
inflexible.

> the mass population haven't. High volume video is actually a
> brain-numbing experience. You stop thinking when you watch it. Do you
> really want to put that on your pc?

I don't own/watch TV. I do watch DVD movies, though. One of the flat panels
doubles as video (my 100$ DVD player has VGA output).
 
> Keen gamers will spend more on the latest graphics card than they
> spent on the computer.

You'll need top line CPU+memory as well as a top line accelerator to achieve
record frames per second. 3d graphics acceleration is useless for game AI.

> The DARPA challenge (and other AI) fails because we don't yet know how
> to write the requisite programs - not for lack of processing power.

Each task has a minumum computational complexity. You can afford being sloppy
by throwing more hardware at it, or vice versa. Up to a point. Realtime
navigation in cluttered environments takes considerably more resources than a
top line PC (more than a cluster of those, actually).

How many generations down the curve do you need to package that into a small
and tough enough footprint on small enough energy budget for automative
applications? 
 
> Yes, if Intel make better processors, then they will appear in all the
> computers we buy, because we won't have any choice. But will the

I presume you mean Intel=AMD in above sentence. It doesn't matter on the long
run.

> majority of users *need* more processing power? I doubt it.

Office users don't. At least not yet. Gamers do, most assuredly. Do you
realize who's driving the sales in high-performance systems?
 
> More bandwidth, more on-line storage and more backup storage, - yes please!

Bandwidth is an infrastructure issue. 1 GBps to each home would be no problem
with current technology, and 10 GBps is almost there. Storage size, not
really, though a TByte RAID looks tight for a HDTV movie library. 

Something solid state with TByte throughput for about the same price, minus
seek latency would be nice, though.

-- 
Eugen* Leitl <a href="http://leitl.org">leitl</a>
______________________________________________________________
ICBM: 48.07078, 11.61144            http://www.leitl.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE
http://moleculardevices.org         http://nanomachines.net
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20040901/1ec0bea1/attachment.bin>