AMD Dual Core for 2005

[TheRod]

<A HREF="http://www.xbitlabs.com/news/cpu/display/20040429105016.html" target="_new">The article</A>

Wow! These chip will be 100% compatible with actual hardware (Socket940).

--
<b><font color=red>GO!</font color=red> <font color=blue>HABS!</font color=blue> <font color=red>GO!</font color=red></b>

 

[P4Man]

I guess this explains the 105W TDP for 90nm.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[flamethrower205]

So this would be dual processor brought to the single proc mobo world...so we can get good mobo performance and pci-x performance? Sick dude. This may be what I've been looking for.

SEX is like math. Add the bed, subtract the clothes, divide the legs, and hope you dont multiply

 

[Mephistopheles]

For some reason, I find it a little hard to believe that actual dual-core chips could be used by single-core-only aware motherboards... But I don't know if I'm knowledgeable enough to say anything about that. Is is conceivable that with the ODMC so much of the actual workings of the mobo is there that they wouldn't truly need any other change except putting a dual-core processor on, say, a current motherboard? Wow.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[Guest]

Captain Obvious says this is patently obvious when one looks at the specifications for the K8 architecture!

THe K8 was designed from the ground up to be multicore capable.



<b><font color=red>Captain Obvious To The Rescue!!! (Obviously)
Captain Obvious: Pointing out the Blindingly Obvious!</font color=red></b>

 

[flamethrower205]

I wonder if we could push this to 3 or 4 cores even? That'd be sick.

SEX is like math. Add the bed, subtract the clothes, divide the legs, and hope you dont multiply

 

[P4Man]

>For some reason, I find it a little hard to believe that
>actual dual-core chips could be used by single-core-only
>aware motherboards

Motherboard doesnt have to be "aware". It only sees 1 CPU (one HTT identifier). Software however, will see 2 CPU's. . Note that the latest dual core PA Risc chip is also just a plug in replacement.

> Is is conceivable that with the ODMC so much of the actual
>workings of the mobo is there that they wouldn't truly need
>any other change except putting a dual-core processor on,
>say, a current motherboard ?

Yes. But even with a more traditional layout its not harder than supporting hyperthreading. I'm fairly certain that future dual core Xeons will also be just a drop in replacement

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[P4Man]

>I wonder if we could push this to 3 or 4 cores even? That'd
>be sick.

Not likely at 90nm. Size and power/heat would be too much to cope with. Maximum power consumption would scale nearly linear with the number of cores. That is also why I am not expecting any multicore netburst chips, they would be too hot. But at 65 and beyond I'm expecting quad cored chips from both AMD (K9) and Intel (dothan and/or IPF derived).

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[TheRod]

The only problem with Dual-Core Athlon 64 : they will be at first only compatible with Socket940 (Opteron). Maybe they will bring Socket939 version with a name like "Athlon FX-2" that would an interesting upgrade for users who have multi-threaded support.

--
Lookin' to fill that <font color=blue>GOD</font color=blue> shape hole!

 

[TheRod]

Yup! this was obvious!

Integrated memory controller, hyper-transport interconnect!

It will be interesting to see the impact on speed, because these CPU will share IMC and will have a theorically full-speed hyper-transport interconnect!

--
Lookin' to fill that <font color=blue>GOD</font color=blue> shape hole!

 

[lhgpoobaa]

But what about bandwidth access to the system ram?
One will have 2 cpu's on the same die accessing the same dual channel DDR ram, unless the pincount is increased for quad channel.

<b>My Computer is so powerful Sauron Desires
it and mortal men covet it, My Precioussssssss!

Regards,
Mr no integrity coward.</b>

 

[P4Man]

Yes, theoretical bandwith per core would be cut in half. however, look at the performance of those dual opteron boards with memory connected to just one instead of both cpu's.. difference is barely measurable for most apps.

In fact, those arima dual opteron configurations should give a fairly good indication of how a single cpu, dual cored opteron would perform; especially now it seems AMD will double the L2 cache (1MB per core). I expected a shared L2.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

 

[Mephistopheles]

I wonder if we could push this to 3 or 4 cores even? That'd be sick.

Well, 3 cores is not likely because it's easier, as far as I know, to do powers of 2 (that's why there are 2-way, 4-way, 8-way and 16-way and beyond systems or whatever).

4 Cores is still very, very far away still. As P4Man said, mainstream solutions won't come from 90nm with quad cores; that's physically very, very inconvenient. Actually, dual cores starts being a truly great idea from 65nm; Montecito, for instance, is a 65nm part. As is Jonah; AMD will also introduce dual core Opterons, but at 90nm - they will probably be expensive to manufacture... But dual-core is quite possible, even on 90nm; quad is not.

And there's always the holy grail of multicore: the 16-cored Itanium/Tukwila (former Tanglewood) processor for 2006/2007 or whatever. God knows if this will materialize; if it does, you could have a quad-processor Itanium platform with 64 processors in a single machine. If scaling and interprocessor communication is performed seamlessly, this should be quite good. Of course, this is just for the server/ultra-high-end segment... And even if you consider the cheaper multicore alternatives - Opteron and Xeon, and possibly A64... well, most of us won't be using a dual-core processor for quite a while.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>

 

[TheRod]

I don't see a problem with this, because of on-die cache and on-die memory controller, the latency and access time will ba fast, even if 2 CPU's share the same memory. Of course, this will impact overall performance, because 2 single core Opteron should be faster than 1 dual-core opteron, but me might be surprised, because dual-core will probably have full speed HT between them, this mean that the data will move with near zero latency between the 2 core, this cousl be advantage in some apps.

--
Lookin' to fill that <font color=blue>GOD</font color=blue> shape hole!

 

[Mephistopheles]

Oh, and into some interesting news: the 2nd generation dual core CPU from IBM (exclusive to servers) is due to be launched officially on Monday, according to <A HREF="http://www.aceshardware.com" target="_new">Ace's</A>. It's the Power5, sucessor to the already dual-core Power4, which is, BTW, a great chip with top floating-point capabilities (exceeded only by Itanium by a big margin). Just in case anyone is interested. Core integration in single dual-core processors has the potential to be much more efficient and fast than dual-processor (single-cored) systems.

<i><font color=red>You never change the existing reality by fighting it. Instead, create a new model that makes the old one obsolete</font color=red> - Buckminster Fuller </i>