AMD will let computer makers pop specialized coprocessors into sockets that otherwise would house an AMD primary processor such as Opteron or Athlon. It's an expansion of the Torrenza initiative which provides a way for others to connect their technology directly to AMD chips, via the company's HyperTransport interface. The program initially let companies plug in their coprocessors via an external connection called "HTX." Now it is licensing the processor socket design as well.
That means a computer maker doesn't have to develop a separate design if it wants to use a coprocessor--a special-purpose chip for handling tasks such as graphics, mathematical calculations or security. "Now an OEM only has to develop one infrastructure," said Marty Seyer, senior vice president for AMD's commercial business.
AMD is pushing HyperTransport as one of its competitive advantages against rival Intel, but Intel processors once again are competitive, and Intel is working on a HyperTransport rival called the "Common System Interface." It's not yet clear whether Intel will let others directly connect chips to its own via CSI.
AMD expects the first socket-based coprocessors to arrive in 2007, spokesman Phil Hughes said. He declined to reveal payment terms for the licensing.
View: Full Article @ CNET News.com
That means a computer maker doesn't have to develop a separate design if it wants to use a coprocessor--a special-purpose chip for handling tasks such as graphics, mathematical calculations or security. "Now an OEM only has to develop one infrastructure," said Marty Seyer, senior vice president for AMD's commercial business.
AMD is pushing HyperTransport as one of its competitive advantages against rival Intel, but Intel processors once again are competitive, and Intel is working on a HyperTransport rival called the "Common System Interface." It's not yet clear whether Intel will let others directly connect chips to its own via CSI.
AMD expects the first socket-based coprocessors to arrive in 2007, spokesman Phil Hughes said. He declined to reveal payment terms for the licensing.

Imagine a board with like 8 sockets, you mix and match what you want, graphics, physics, cpu, etc.....That would be pretty cool, and great for customizability. But is that the direction things are going? Or I am I looking at this wrong?
If it did use system RAM, that would be ideal. You would get such a huge speed increase by such a monolithic view of memory.
Yeah!!! Let's splinter the market even more!!!
Good job AMD.
And it has the disadvantage of having to rely on mobo circutry and on-board chipsets to enable said modular chips. Whereas with expansion slots, you can build a card that has whatever features you want. So, in other words, this is an attempt by AMD to sucker the discrete-card makers into giving up control of what they make and what's on what they make. Don't fall for it.
Actually to the guy that says it's not a coprosessor, that actually just what it is. Well technically you coudl use the port for anythign that requires high speed connection to the cpu and memory, but that's not what it's designed for.
it IS designed for a coprocessor. and no it's not meant for regular consumers.and I see it unlikely that any grpahics card makers will make cards that fit these slots, unless they're building systems that are completely integrated in a single non upgradeable box(Essentially the future of home compputers anyway but..)
This is meant for indsutry and insititions that require special purpose processing units and such.
Medical research facilities can have a special coprocessor built especially for their needs. Large studios who do 3D rendered movies can add on coprocessors designed especially to do rendering. The only thing I see this being possibly used for outside of specialist industries is for physics coprocessors, and frankly for that, this is a much better solution that those useless Ageia cards.
And even so coprocessors aren't a step backwards, whent he math corprocessor was added as standard, that was ebcause everyoen pretty much needed it anyway, why should regular CPU's be stacked with instruciton sets that are never used. RISC cpu's are faster than CISC, but dont' have enoguh instructions, so today most cpu's are a RISC/CISC hybrids. Now if we added on all the instruction wanted and needed by different industries we'd have a cpu so clogged with instruction it would be very very slow.
Let's take a peak at the last true consumer RISC CPU, the older Motorola Apple CPU and early PowerPC cpu's used by Apple. These where CPU's designed around one simply thing, converting color tables(not as easy as it sounds but you need some basic color theory to understand what it really is). This mean that for certain tasks, like photoshop and photoshop filters these offered vastly higher performance per cycle than an x86 CISC cpu. on the other hand, the lack of other more generic instructions meant they actually lagged behind in most other areas of computing, like 3D, plain math calulaitons and database stuff and things. Hence why apple allways used photoshop to show of how much faster they where, and why Apple was (and to some degree still is, though it's a legacy thing) used by so many designers and graphics workers.
Now through this interface, specialist industries can add on RISC coprocessors designed to do only one thign, and do it very well. removing the need to design whole systems, instead they can just use regular AMD computers(or Intel with CSI) with regular x86 kernel OS' liek windows or Linux, not needing to write specil kernels and crap. and just add on drivers for the corpocessor.
We're not movign back in time where everyone had to buy a math coprocessor for their computers, and I doubt many if any consumer mobo's will even have a htx/csi port. so quite yer whining
That's exactly what i was thinking as i was reading the article...
and i love the people who complain about AMD doing it, making it "splinter the market" when the last few paragraphs say that Intel is making their version of it! hahah gotta love people who don't read the whole article before they go off on something...
If AMD is letting third party computer makers put third party processors into AM3 sockets, would that not leave room for Intel to make a special Core 2 Duo processor with Conroe and a special memory controller chip in the same package and sell it as a replacement for AMD Opteron and Athlon 64 processors?
If that is true, then AMD might be making arrangements for its own funeral; although, if Intel was to do that, it would have to start making chipsets for AMD processors, because everyone would switch to AMD's partners' chipsets, because of the choice that it would give them of either Intel or AMD, and if everyone switches to AMD's partners chipsets, Intel's chipset business will lose the majority of its marketshare; thus, forcing Intel to make chipsets for AMD processors.
I really think that this could enable Intel to competitively drive AMD out of business; that is, if Intel decides to start making processors for AM3.
Imagine the the benefits having CPU making direct integration with graphics processing unit, disk controller, network controller and sound processing unit without having to go through some type of a "bus" system...That's the future of computing...
I think its great, as many people enjoy modifying there mobo to accelerate over the suggested requirements.
When you closely look at any mobo, don't you wish you could upgrade any chips to newer or better.
Sofware makers are now focusing at many of the mobo capabilities thus improving the end user experience.
In conclusion, we are now not limited what is soldered on our Motherboards, I think its a great lead away from Intel.
as to the two ding dong about going back in to time, Any new process must follow thru compliances from many IT international standards. It would be impossible to fall back.
for the record, reading todays forum and past editorials of the ding dongs, please your better off to just read the forums here than post an ignorant don't know what your talking about negative statements about any issues.
as to Amiga, it was envision to work quickly as a good idea, but a bad longterm solution to todays advancements, that is why Windows Prospers over Amiga.
The ding dongs on here, you guys really need to spend more time with your girlfriends than try to be a mister know it all that you have no clues. Todays IT forums are way past your intelligence.
Underlining your insults directed toward other members is a good idea. It helps the mods easily see who is calling whom names.
This is just a guess, but I'm assuming this will add an extra set of instructions to the processor. Think MMX or SSE2. A company will license the techonology and build their special coprocessor which will add a new instruction set for programmers to use. Processor intensive things would be the target here. Think things like physics, or maybe an encryption addition for computing hashes and other epensive calculations. The companies would also create libraries to use their new functions, or allow programmers to call the functions directly, with assembly, on their own.
The idea of Intel being able to replace an AMD proc with intel is silly, although I will agree the article makes it slightly ambiguous at first glance.
As for profitability - they should make licensing this technology as cheap as possible, even free. The coprocessor will be a complement (a product sold in conjunction) with their own mobo chipsets and processors. The cheaper the complement (in this case, coprocessors), the more demand you have for your product (chipsets and processors). It would encourage companies to design for, and use, HT. In short: cheaper HT licensing is equal to more HT licensees. The more coprocessors you have the more demand for your chipsets and processors you have. Since the coprocessors will be highly specialized, you'd be selling high-end chipsets and processors for use with the coprocessors.
Again, this is only a guess.
Commenting has either been disabled on this article or you are not logged in. Click here to login or register, its free!
Note: Anonymous commenting is disabled in order to keep the quality of responses to a high standard.