Fog Creek Software
Discussion Board




Chip progress is incremental, not revolutionary

http://nytimes.com/2004/02/11/technology/11CND-CHIP.html?hp

I suspect most of them are fake advertising with a rats chance of getting into the market place. Anyone heard of IBM's "Advances" that keep coming or the single atom switch that was advertised long back?. There was another item about some harddisk that stored data like punches that could store about anything.

I bet that this is the last time you will hear of it. The only major innovations that have taken place in the computer industry since 1980 has been the CD and DVD. Everything else has been incremental. IBM is the worst in this kind of fake advertising. Most of their "revolutions" never get past the R&D Stage.

Karthik
Wednesday, February 11, 2004

Someone ought to make a blog of these and hold these companies to account. I used to be extremely interested in this sort of thing. Later, i realised none of them actually get sold.

Karthik
Wednesday, February 11, 2004

In the future, computing power will be sold by the feet. With integration going the way it is, and distribution systems (read: Internet) working the way it does, it won't be long before flexible magnetic memory+flexible cpus/controllers+fiberoptics with easy variable length tap in and tap outs allow vats of cables and coins that compute. And once we hit that point, production will REALLY ramp up due to lack of moving parts and less administration fudge ups.

Li-fan Chen
Wednesday, February 11, 2004

This is how variable length tap in tap out works.

Say you have a fiberoptic cable.. it has 500 full length threads.. + 5000 variable length threads..

the design goes like this.. the 500 threads goes from one end to the other

the other 5000 makes end points possible anywhere between end A and end B...

some are 5 feet long hubs.. some are 10 feet long paths.. some are 100 feet long paths..  and depending on computational model it will probably distribute like this:

65% are hubs 20% are 10 feet long paths.. and rest are longer paths

that way you build a network that can digest streams of jobs any given inch of computational unit can intercept/forward.

you heard it here first

Li-fan Chen
Wednesday, February 11, 2004

Not right on point, but something a prof of mine used to say...

Definitions:

Statistics: numbers.

Science: using Statisitics to make a point.

Tech: Science you can use.

High Tech: Tech you can make money on.

Nigel
Wednesday, February 11, 2004

I saw this article too -- I haven't read it carefully, but isn't it about an optical switch?

I remember back when I was interning at Xerox PARC that was the holy grail for a few researchers.  I am not a hardware guy, but the idea is that you don't have to convert from optical -> electrical -> optical for routing optical signals, you can just keep it all optical if you have an optical switch, and that speeds things up by orders of magnitude.

So I guess the implication is that bandwidth is limited by switches and not by the capacity of the line itself.

If that is the case, then it might not be as overhyped as it sounded.  The NYT is generally less subsceptible to hype than most newspapers -- notice that they are careful to put "Intel says" in the headline.

Andy
Wednesday, February 11, 2004

I think the original poster misses the point. IBM brought a lot of innovations on the market. The fact you don't see them, doesn't mean they are not used. Ever heard of magneto-resistive heads? They are the sole reason you enjoy now a 120GB hard drive smaller then a closet.

The optical technologies in the article refer to the possibility of implementing optical switches on the same chip with the CPU and using the same technological process. This would be a huge advancement since transmitting microwaves (1 GHz + is microwave domain) through the CPU bus is a big physical constraint on the present CPU design. An optical switch might replace the CPU bus and the FSB completely in the next few years

coresi
Thursday, February 12, 2004

*  Recent Topics

*  Fog Creek Home