Sun finds way to make computers 100 times faster

[Nottingham]

http://straitstimes.com/latest/story/0,4390,211280,00.html?

Sun finds way to make computers 100 times faster
SAN FRANCISCO -- US high-tech giant Sun Microsystems on Tuesday unveiled what it heralded as revolutionary new technology that could make computers 100 times faster and significantly smaller.

The company has discovered a way to connect a computer's chips face-to-face rather than linking them through the web of fine wiring currently laid down on circuit boards, a Sun official said.
By eliminating the wiring, the discovery could speed up data movement by as much as 100 times while reducing power usage, he added.

It could also eventually bring an end to the use of printed circuit boards, the wire-laden nerve network which takes up much of the room in electronic products.
Sun, a leading computer networking specialist based in Santa Clara, California, revealed its discovery in a paper presented to a conference on integrated circuits in San Jose on Tuesday.
The new technology's potential impact on computing is immense, said Mr John Gustafson, a senior official at Sun's high productivity computing systems division.


more....

Awesome discovery! :bounce:

[wtmusic]

1. 'Sun's researchers are wrestling with problems of excessive heat'

This has been tried before, and it was the same problem; maybe they have a new angle but I'll believe it when I see it

[republicansareevil]

19. Well of course the heat is excessive on the sun!

:P

[sam sarrha]

27. I worked on F22 fluid cooled dual sided boards that ran over 180*

they were really neat, about the size of a piece of toast. they had snap on coolent lines.

[loudsue]

2. Kewl!!! I'm ready for fast!!

Good news from Sun. I'll be the stock is up.

[Braden]

3. cool

how can you connect all of the chips together I wonder? You have a processor, logic, memory, specific chipsets for graphics or networking, but how can they all interconnect without a circuit board?

SUNW trading at $4.09 per share now, a five year low.

[lcordero]

6. It has gained a dollar in the last year

http://finance.yahoo.com/q/bc?s=SUNW&t=1y

[gkdmaths]

9. you dont connect the "chips"

rather you make a "chip" with all components in it. for instance, by throwing enough room for all the addresses for, say 512 Megs mem (actually, if you integrated it, you'd need considerably less memory to run significant applications), directly into the architecture of the processor, you eliminate the bus that slows the PC down the most.

voila

:shrug:

perhaps

[Atlant]

12. There's a difference between PCs and "Enterprise-Scale" systems.

> you dont connect the "chips" rather you make a "chip" with all
> components in it. for instance, by throwing enough room for all
> the addresses for, say 512 Megs mem.

You're thinking about PC-scale systems.

I'm sorry, but you can't build "Enterprise-Scale" servers that
contain all the system memory "on-chip". We're talking about
systems that share MANY GIGABYTES of main memory among
tens or hundreds of processors. Even at the rate that Moore's
Law delivers transistors, it'll be a long time before you can
put all that on a single chip.

Interconnects are a fact of life in large systems, and that's
where some of the most-exciting Computer Engineering is going
on. (Most systems now use some sort of "switch fabric" as their
backbone interconnect with many simultaneous data transmission
paths.)

Atlant

[Noordam]

21. But if you onboard 512MEG local to each processor

and use that 512Meg CACHE you start building a very very fast multiple processor system.

[Atlant]

24. Yeah, but NUMA technology isn't that good yet.

> But if you onboard 512MEG local to each processor and use that 512Meg CACHE
> you start building a very very fast multiple processor system.

Yeah, but last I knew, NUMA (Non-Uniform Memory Architecture)
systems still have a lot of constraints compared to a system with
a uniform memory access model or a switch-fabric that minimizes
the "NUMA Penalty". For some programs (with small, not-very-
shared data sets), it's great, but for other programs (with terabytes
of shared data), it's not very effective.

I'm sure you're right in the long run, but there's still a lot of work
to do.

Atlant

[Caution]

4. yep...remove all forms of java

(sorry but I freakin hate Sun :)

[tridim]

11. Yep, Sun sucks

Their products are criminally overpriced. Thankfully businesses are finally figuring out what the market has known for awhile. My last interviewer said they recently replaced all their Sun boxes with Linux PCs and will save millions in the first year.

[Noordam]

22. I love Sun and Java

you can take your .NET and deposit it up * .............. :evilgrin:

[Atlant]

5. The article didn't contain enough data to decide if there's any...

The article (and related articles) didn't contain enough data to
decide if there's any "there" there. The main packaging issue these
days ain't the copper versus the ether, it's the speed of light and
the heat, neither of which can possibly be addressed by this reputed
Sun breakthrough.

We'll see what develops.

Personally, I'd bet on optical (photonic) interconnects trumping
all electronic interconnects in the near future; optical interconnects
have the very desirable property that they can travel in three-
dimensional free-space without guideways or crosstalk, something
electrons or electrostatic charge doesn't readily do. This would allow
VERY dense interconnect topologies.

(Anybody else remember those Raytheon "electronic blocks" that
contained electronic components with electrical contacts on the
four sides of the blocks? They stuck together by magnets, IIRC,
and you could stick a bunch of them together to build a simple
circuit. I thought of them immediately when I read this press
release.)

Atlant

[Nottingham]

8. Very Interesting! So what you guys are saying is

The Future in computers still has a ways to go! :bounce:

[gristy]

7. It's called multi-chip modules.

This is not revolutionary. A small evolutionary step, if that. There's really nothing here.

And here's a good little joke: It could also eventually bring an end to the use of printed circuit boards

So, the integrated circuits are going to float in the air? Connectors are going to be mounted directly on the integrated circuit? DUers excepted, this kind of press gets mileage because people are so clueless.

[Nottingham]

10. Well it looks good on paper! Baaawaahaa! Thanks people

if your not in the computer industry this would look Awesome

when in reality ho hum!!

DuERS rock!

Hold up on buying that Sun stock people! :bounce:

[SaintLouisBlues]

13. PC boards aren't going anywhere

Even if they manage to mimimize the PC board content of computers, there are still tons of lower-tech applications where PC boards will continue to dominate.

That being said, the domestic PC board industry has gone from around 4000 companies fifteen years ago to a few hundred today.

Meanwhile, the Chinese PC board industry is going great guns.

[demdave]

16. This is because the Chinese can dump their toxic bypreoducts in the rivers

The expense and difficulty of disposing of the saturated etchants or reclaiming the metals back out of them has killed this industry. Plating solutions and used etchants, not to mention off-gassing during electroplating, are all nasty products. Don't get me started on the photoresists.

[SaintLouisBlues]

26. Making boards is a nasty enterprise alright

But the Chinese are also taking over because so many circuits are now
manufactured over there, it makes it even cheaper to buy boards and components locally.

Outsourcing trickles down in a big way.

[Atlant]

17. Actually, Sun seems to be discussing something different than MCMs.

(Based on the scanty info that I've read) Sun seems to be proposing
the use of capacitive coupling (via near-field effects) over narrow
serial "busses" rather than ordinary low-resistance wiring (over
what are usually broad parallel busses but are also starting to shift
in the direction of narrower, self-clocking serial busses).

MCMs, by comparison, are, as you observe, "old hat". Funny
how fast the hats get old in this business, isn't it? :-)

Atlant

(Who, a long time ago, wrote a chapter of Joseph Di Giacomo's
"Digital Bus Handbook":
http://www.amazon.com/exec/obidos/ASIN/0070169233/ref%3Dnosim/websleaderboo-20/102-2047211-2784108#product-details
)

[gristy]

25. That's what I design. Been doing it since '95

but are also starting to shift in the direction of narrower, self-clocking serial busses).

[demdave]

14. Here is a much more complete discription, especially page 2

http://www.nytimes.com/2003/09/22/technology/22SUN.html

In essence what they sre trying to do is put an entire computer on one silicon wafer. The draw back is in the defect rate on the wafer. X number out of 1 million connections fail. By being able to put chip edge to chip edge, they can take only non-defective chips and reconstruct the wafer. The breakthru is the mechanical precision placement of the wafer segments (chips). It would have to be accurate to about 1 or 2 microns. That is a breakthough.


The heat problem would still have to be addressed but would not include the added heat produced by having the large (relatively) 100 square micron bonding pads for off chip connections.

[Atlant]

20. Somebody whisper "Trilogy Systems" in their ear...

Thanks!

> By being able to put chip edge to chip edge, they can take only
> non-defective chips and reconstruct the wafer.

Before they go applying for their patents, somebody better mention
Gene Amdahl's "Trilogy Systems" to them; that "reconstructed wafer"
was the basic concept behind what Trilogy was trying to do:

http://actscorp.com/acts/amdahl.htm

> In 1980, Dr. Amdahl co-founded Trilogy Systems Corporation to produce
> fault tolerant wafer-scale chips and a high-performance CPU. The
> concepts were proven but the cost projections were excessive, so
> Trilogy acquired Elxsi in 1985 for its principal computer system entry.

Of course, the capacitive edge-to-edge interconnect may be new.

On edit:

Ahh, here it is at the bottom of the NYT article:

> All of them have failed and several have collapsed in spectacular
> fashion. Gene M. Amdahl, the designer of I.B.M.'s 360 mainframe
> computer, founded Trilogy Systems in 1980 to build an advanced
> mainframe computer based on wafer-scale technology. He was able to
> raise $279 million from computer partners, venture capitalists and
> a public offering, before going under.


Atlant

[davidinalameda]

15. yeah! I can get my spam and junk mail 100 times faster now!

whoo hoo!!

[rocktivity]

18. How sad--that's going to result in a LOT of layoffs

...oh, you mean make the computers WORK 100 times faster? Never mind.


rocknation

[srpantalonas]

23. Sun outsources TONS of work, including MY software from my old company!

http://www.CrystleForSenate.com

Also, the L-1 Visas are killing technology opportunities. The recovery isn't jobless--it's just jobless for American workers.