AMD with SMT (hyperthreading)?

[m0ti]

From The Inquirer:
OVER AT THE US trademark and patent office, there's a slightly elderly patent successfully filed by AMD back in 1999 which indicates that the firm could provide hyperthreading for its processors, if it should so wish.

The patent, numbered 5,944,816, entitled "Microprocessor configured to execute multiple threads including interrupt service routines", outlines how a CPU can execute multiple threads concurrently.

It specifically says that in one test, a CPU executed at least two threads concurrently.

The patent builds on a number of technical documents on simultaneous multithreading, including an Intel document from 1994 called Multiprocessor Specification.

So will we see a hyperthreaded Clawhammer (Athlon64)? Well the patent seems to suggest that's not outside the bounds of possibility.

Down and dirty with the details: link

If I'm reading it right it doesn't talk about true SMT. It only discusses running a thread and multiple ISRs (Interrupt Service Routines) concurrently and not multiple threads in general. I guess The Inquirer is going a bit far in predictions for HyperThreading.

Though, conceptually, it is good to see that AMD has some early form of HyperThreading. I'm practically convinced that we'll see them with SMT capabilities by K9 (don't know what they've got planned for revisions of hammer, but K9 will almost definitely include this, IMO).

What do the uber (and not so uber) geeks think?

 

[Mday]

this smt rumor has been around for at least a year or 2. so far the only thing amd has said is that their intial 64bit offerings are single core. my initial thoughts were possibilities of multicore processors, like what ibm has.

as for hyperthreading, intel owns that name.

 

[m0ti]

HyperThreading is, of course, only a marketing name.

Hence the SMT. There's been more speculation about multiple cores per die than SMT, granted, but it probably will show up in the roadmaps at some point.

 

[grant2]

I'm certain AMD will include SMT eventually, unless intel's performance gains turn out to be hocus pocus.

 

[Howard]

The gains while using HT in some applications are definitely NOT hocus pocus.

 

[NovaTone]

I was at the HardOCP / AMD event in dallas yesterday, and from what I can gather, AMD and Intel have differing methodologies for their upcoming processors. For Intel, they're banking on hyperthreading, while AMD is putting their money on an ondie memory controller. If hyperthreading picks up popularity, Intel lucks out, and AMD will definitely have to incorporate hyperthreading into their processors.

 

[PentiumIV]

While SMT sounds relatively simple to implement, it is REAL HARD to make it work,
especially for out-of-order processor. It is quite tough to validate all the boundary
conditions and to ensure sane performance and functional correctness. So,
I have serious doubts that AMD will do SMT on some proliferation of K8, since it requires
100% new core, or otherwise it may have some undesireable impact on frequency/IPC.

 

[Mday]

right now, amd is really depending on hammer for it's continuation into future ventures. if amd does have smt, any implementation, then they are not going to announce it "yet". then again, if they dont have that tech "yet" they wont be making any announcements. i am not the kind to rummage through patent filings and claims to find out technology that companies have. and i doubt i would be able to make heads or tails out of it if i do =\

--

oh god gregory, of all the ... your nick is p4 =\