Will overclocking days be over.. look here
 

It looks they have invented a new coprocessor capable of outputting 25.6 gigaflops. According to this company "the ClearSpeed's 32-bit CS301 coprocessor runs at only 200MHz but outputs up to 25.6 gigaflops per processor. The company's chief designers envision the chip perched on a PCI daughtercard, assisting the main CPU with computation-intensive parallel tasks, such as those used in the biotechnology and scientific communities. "


More info here

I would like to get my hands on one for my DF cruncher. he he

One?... from the sounds of it you can have serveral daughter CPUs :D.

Or it could just be another hoax - like that went round a few years back with the PCI processing card for SETI@home.

~ Boli

A) overclocking will never die, no matter what a company can put out, people will always want more.
B) I shudder thinking about good interaction with a cpu at the level we demand over a pci limited bus.
C) I will believe it when I see it.

Interesting.

Given that PCI is basically a dead bus, this could be cool. Give each of them a PCI-X channel or something of that nautre.


Or you could adapt it to the the new Opteron architecture. Put a northbridge line on each processor and have a few co-processors for each processor. that would make one hell of a work horse and still find a way to not overheat the planet with it.

=)

Now we just need some software to use it all.

Need some software for it? Distributed folding at your service.

Until they find a way to tie something like this directly into a wide bus such as hypertransport and provide an environment that can do intelligent multiprocessing with these (i.e., farming out threads that could be more efficiently run in such an environment but keeping threads that are memory hungry local), I don't think many people will be too terribly interested in the enthusiast community. Give me cheap and easy SMP that is intelligent, and I'd be game, though.

25Gflops for only 3W of power?

"Soon to be in prototype?"

Sounds like vaporware.

if this is true...well I'm glad. Unforuntately it'll prolly cost a problem...

As mentioned, what about the amount of heat potentially produced?

Power is not just a function of speed. Any array of transistors that pumps out 25 gigaflops of data is going to produce more heat than even a watercooler can dissipate. Transistors switch, they generate heat.
In order to increase performance and maintain a low clockspeed a processor has to be extremely parallel, which means MORE transistors. More transistors = more heat. Regardless of if each transistor radiates 1W of power or .1W of power, performance is always directly related to heat.
For example, look at the Cyrix III cpu, VIA boasts that it can run without a heatsink, without a fan in fact, but most people would buy a 900mhz Celeron over a 900mhz Cyrix III simply because it blows.
Sounds like a bunch of crap to me.
-Zoson

Maybe they are a sister company of BitBoyz :) hehehe
The Vaporware kings

They are estimating costs of only $16K per expansion card. Who's first in line?

Hummmm, New car.........or.........expansion card....... :D

Keep in mind that it doesn't actually exist yet ;). The "will enter prototype soon" thing says that maybe it won't pan out as planned. I certainly hope that it does though ;).

it'll head in the same direction as Transmeta...

It would be possible to create a low-power, low-cost, high-flops processor, but it would end up being a VERY specialized processor. A cpu that only has to support a single instruction or two can be very good at it (them) without needing millions of transistors and tons of power. If it was a general-purpose processor like an Athlon or Pentium they likely wouldn't be calling it a co-processor or putting it on expansion cards. It'll be a very specialized processor for specialized tasks.

Judging power by required heatsink size, compare an Athlon64 FX or similar chip to say a Radeon9800Pro. The radeon is WAY WAY faster at doing the few instructions that are required to render a complex 3D scene, but (assuming it could) you probably wouldn't want to try to run windows on it.

When the word "coprocessor" is used, it usually means a "Math Coprocesssor", thus they do tend to run easily in any system with total disregard of the operating system

Do you remember when the first math coprocessor came around for the 33 Mhz cpus? I do, and yes I am that old, but the point is that in the begining It was a separate chip, much like a bios chip, and later it was incorporated into the cpu. Hell, I even remember "software" version of math coprocessing for those that did not have the blessed lil math coproc.

Math coprocessors do not require much interaction with anything else running in a computer. As their name implies they are only used for math calculations which are very simple give-an-take black box operations from the standpoint of any bus/cpu.

That was the difference between a DX and an SX if I remember correctly. I belive the DX had the co-processor built in and the SX required the add on chip to become a DX, could be wrong though. I still have a 25mhz DX board around here somewhere. Had a 16mhz SX at one time but I think I gave it away.

heh... i wonder how long it would take those old 286 12mhz machines we got rid of not that long ago to start win xp pro :)