[JonWoodruff]

VansHardware claims that AMD made one of the most exciting announcements of WinHEC behind closed doors.

AFAIK the biggest hint we've seen concerning what that might be is that Barton is totally different than what we thought it would be. (non-SoI and 512k of cache).

Could Barton be even more revoloutionary?

Penstarsys.com suggests that Barton may have the 64bit x86-64 instructions added to it, though still remaining socket-A with the 64bit instructions implemented in 32bit pipelines.

Feasable? It would be simple to do, and seems to be exactly what Intel might do with Prescot, according to chip-architect.com. These hybrid implementations wouldn't be as fast as the hammer, but would do a lot toward driving mainstream acceptance of x86-64 without having to cheapen the full-fledged hammer. Barton could serve as the fully-featured "MX" version of the hammer for a long time to come.

[Saem]

I'm not sure how feasible this is. An architectural conversion doesn't sound cheap. They're dramatically changing the underlying architecture if they add x86-64, the additional registers, the changes in the decoders -very complex- the changes to the ALUs and FPU that would be needed. Then there might be a significant amount of time needed to reogranize the core and work out any new critical paths that arise.

[pascal]

Lets say it is true, then one possibility is a socket A clawhammer core (without onchip memory controller) to get the socket A installed base market.

[pascal]

AMD Roadmap sugesting Barton with 512kb l2 cache: http://www.amd.com/us-en/Corporate/V...04_608,00.html
It is old link I got from Acer´s.

[Saem]

Lets say it is true, then one possibility is a socket A clawhammer core (without onchip memory controller) to get the socket A installed base market.

Possibly, but wouldn't this be a bad business move. Mind you the Barton even if it does have x86-64 capabilities will be significantly slower, lets face it nearlly all of the performance improvements are coming from the memory controller, when comparing on the same binary. I believe there might be a slight increase from improved branch prediction.

But, back to Barton have x86-64 being a bad move, if they do this, they'll give the market willing to buy an AMD a reason not to bother with ClawHammer and to just keep their money and go with a cheap barton upgrade. I believe AMD will nod bring x86-64 to the Barton. If I were them I'd switch all the k7 production to their joint venture with TSMC ASAP and then go nuts on ClawHammer and Hammer chipset prodcution. If they can aggressively produce them, their manufacturing costs and thanks to their higher sales price a la marketing ratings will have them producing a significant amount of money, while still being able to fulfill their old K7 order obligations.

[pascal]

Maybe the idea is give the x86-64 compatbility without the performance. Then most people will still prefer Hammer, but people on budget with legacy hardware could buy a Barton. This could generate a very large installed base of x86-64 capable computers and incentive the software migration/developement.

But I agree that develop an x86-64 compatbility is not an ease task.

[JonWoodruff]

I agree that there wouldn't be a big performance improvment by making Barton x86-64 compatable, but I wonder how difficult adding x86-64 compatablility would be. 64-bit operations are already executed by the Athlon core, but require several instructions. It seems (relativly) simple to add the new instructions to the decode stage, so that the new 64-bit instruction is decoded into a series of micro-ops to perform the 64-bit opperation using 32-bit pipelines. Creating or emulating the extra registers may be a bigger problem.

The speed improvement over 32-bit software would be small, but compatability would be there. Compatability is very important in pushing a new standard, so AMD's choice is to either quickly distribute the Clawhammer core from top to bottom (which they may very well do) or to convert the Athlon core to be x86-64 compatable and command a premium for a real Hammer (which will really be much faster).

AMD is shifting Athlon production to UMC while moving Dresden to Hammer. AMD would like to offload capacity to UMC, so we know that AMD expects Athlon production at UMC to continue for a while. Would AMD continue to sell the Athlon without x86-64 support, even as a value chip? To me, an Athlon 64 in the value segment produced at UMC sucking up the ~20% of AMD's capacity that the Duron now occupies with the Hammer in the mainstream makes a lot of sense. Otherwise, either their old 32-bit core will remain a thorn in their side for a long time, or all the infrastructure support they've accumulated for Socket-A will be abandoned very quickly. The x86-64 conversion seems like the best option, if it's not too difficult.

[Saem]

The speed improvement over 32-bit software would be small, but compatability would be there.

Actually, there would be a significant improvment, I think you're glossing over the added registers. They give a significant performance boost.

AMD's choice is to either quickly distribute the Clawhammer core from top to bottom (which they may very well do) or to convert the Athlon core to be x86-64 compatable and command a premium for a real Hammer (which will really be much faster).

You'll have to forgive any precived snobery, but I don't believe AMD is capable -doesn't have the necessary resources- of making a conversion. I think it's been established that it would be a major redesign. New decoders -they've very compled- update all the execution units. So I don't think x86-64 capability will be present in Barton. UNLESS, Barton is a Hammer without 512KB cache and the usual K7 bus interface.

With all that said, I believe AMD will have an easier time driving yields up and pushing ClawHammer prices down, then that should tie them over until Sledge takes the top end and Hammer takes the middle and their old stock and K7s take the bottom.

[Saem]

The speed improvement over 32-bit software would be small, but compatability would be there.

Actually, there would be a significant improvment, I think you're glossing over the added registers. They give a significant performance boost.

AMD's choice is to either quickly distribute the Clawhammer core from top to bottom (which they may very well do) or to convert the Athlon core to be x86-64 compatable and command a premium for a real Hammer (which will really be much faster).

You'll have to forgive any precived snobery, but I don't believe AMD is capable -doesn't have the necessary resources- of making a conversion. I think it's been established that it would be a major redesign. New decoders -they've very compled- update all the execution units. So I don't think x86-64 capability will be present in Barton. UNLESS, Barton is a Hammer without 512KB cache and the usual K7 bus interface.

With all that said, I believe AMD will have an easier time driving yields up and pushing ClawHammer prices down, then that should tie them over until Sledge takes the top end and Hammer takes the middle and their old stock and K7s take the bottom.