AMD Quad Core Discussion & Editorial

[The-One]

And what do you suggest instead of that approach?

Cause you just got me curious. Is he a talker, or a smart one?

Your choice to disclose.

He has a point but it's the easiest way to boost performance at this point in time.

But a bigger more power consuming product ain't overall better,it's just better performance.

 

[Enduring_Warrior]

I am reminded of Torrenza. This is the first step. What would most of you guys do if you only had one PCI slot? It'd suck. I know. Becouse we got sound cards and Ethernet cards, and other devices that need to be plugged in...

Soon real soon Torrenza will be on us.

May i ask what Torrenza is?

Cant seem to find info on it.

Thx.

 

[duby229]

One of the things that was mentioned about the second socket in the 4x4 is the application specific co-processor possibility.

In the enthusiast arena, there is only really two application specific co-processor necessity, graphics and physics.

It is not hard to imagine that ATi can support and design a graphics chip as a co-processor, but will Nvidia embrace it as well? Hard to say since PCI-express is right now the industry standard.

Who will design a physics co-processor for AMD remains to be seen... I think it is too far into the future to speculate about these possibilities since third party design will only be viable if the 4x4 platform is an overwhelming success (which is also remain to be seen).

For now, I would say 4x4 is a quad core alternative to Intel's Kentsfield in the immediate future.

It's all grasping at straws the truth is that even socket 940 has had several released coprocessors. And it has already been stated by several companies that it is there design goal to validate for socket AM2. Think about what an FP coprocessor can do, if it was designed to execute SSE instructions? Or what about AI coprocessors? Or maybe Physics? Or Graphics? What about the possibility of HTX? Whast do we do when new NPU's are validated for it? Or Audio Processors? Or RAID controllers? Or Fiber interconnects?

Does Intel have anything that can even come close? The quick answer is HELL NO! 4x4 is the first step in the right direction. You got to make a first step some time, and this is it.

 

[Lolento]

And what do you suggest instead of that approach?

Cause you just got me curious. Is he a talker, or a smart one?

Your choice to disclose.

I think MCM is a much more elegent solution than 4x4, but that's my opinion.

With 4x4, for AMD you literally just pull a reference server design.

MCM will be more involved, encompassing two small dual core dice, better line/space control in substrate, better wafer test coverage (because 1 defect die rejects the module)...

 

[Lolento]

It's all grasping at straws the truth is that even socket 940 has had several released coprocessors. And it has already been stated by several companies that it is there design goal to validate for socket AM2. Think about what an FP coprocessor can do, if it was designed to execute SSE instructions? Or what about AI coprocessors? Or maybe Physics? Or Graphics? What about the possibility of HTX? Whast do we do when new NPU's are validated for it? Or Audio Processors? Or RAID controllers? Or Fiber interconnects?

Does Intel have anything that can even come close? The quick answer is HELL NO! 4x4 is the first step in the right direction. You got to make a first step some time, and this is it.

I'm not questioning the possibilities of what can be coprocessors. On paper, it sounds great.

But there is a reason why Ageia didn't design PhysX with PCI-express interconnect. Using a PCI slot would give them a 100% market coverage. If they design a PCI-Express card, they would limit their market to SLI and Xfire configs; on top of that, they would compete with ATi and Nvidia for that last slot....

Similar argument can be applied to the second socket in the 4x4 platform for third party support.

So, like I said before, 4x4 can't be taken as much more than a quad core alternative at the present time. (Unless, of course, some third party design houses suddenly come out of the woods and claim some design in progress)

 

[Jason711]

they would compete with ATi and Nvidia for that last slot....

that isnt true, the lesser pci-e slots would be more than sufficent.

 

[wizzackr]

You said two things:

1) low-cost DC(dual-core) workstations.

bugger, my bad - I was talking about DCC and lost the second C somewhere on the way , hence the confusion.

 

[visaris]

I think MCM is a much more elegent solution than 4x4, but that's my opinion.

With 4x4, for AMD you literally just pull a reference server design.

MCM will be more involved, encompassing two small dual core dice, better line/space control in substrate, better wafer test coverage (because 1 defect die rejects the module)...

While a MCM may be smaller and may be more "elegent" in some respects, the quad-core MCM doesn't really perform as well as a two-socket dual-core rig. This holds for both AMD and Intel. Clovertown presents two loads on the FSB and the speed needs to be reduced to support it. Also, when there are two chips on one FSB, each chip only gets half the (reduced) bandwidth. MCM sounds pretty, but once you implement it, you realize that there are many down sides.

Using two sockets removes the double-loaded bus issue, removes the bus contention (bensley has DIB). In the case of AMD, using two sockets actually adds a memory controller.

It may seem like 4x4 with two sockets is some kind of a hack (and in a way it is), but when you compare the performance of a dual-socket to the performance of using the exact same chips in a MCM, you can see that the MCM has some disadvantages.

 

[(cf)Eclipse]

I think MCM is a much more elegent solution than 4x4, but that's my opinion.

With 4x4, for AMD you literally just pull a reference server design.

MCM will be more involved, encompassing two small dual core dice, better line/space control in substrate, better wafer test coverage (because 1 defect die rejects the module)...

but with 4x4, you have the option of dual quad core down the road

edit: AND 2x the memory bandwidth of a single socket solution too

 

[quadnad]

Question: does anyone know if there will be four memory banks, two per socket (i.e. one stick per bank for dual channel) or will there be four per socket (meaning I've got no clue how that would work).

 

[(cf)Eclipse]

Question: does anyone know if there will be four memory banks, two per socket (i.e. one stick per bank for dual channel) or will there be four per socket (meaning I've got no clue how that would work).

quad channel is not happening anytime soon. too impractical. plus, you would have to completely change the motherboard for it to work anyway. we're sticking with dual channel for now, and for the forseeable future

though you said 2 per socket.. you mean for 4x4? it should be NUMA like the opterons... each socket is dual channel, so each cpu has dual channel memory access for a grand total of 4 channels, but that's a little different than quad channel

 

[RPhArrow]

As a "normal" desktop user with one exception, I also do not necessarily "need" quad core machines. However, my one exception is folding... and is also the only reason I recently purchased an Intel E6400 Allendale dual core CPU. Was the payoff sufficient for my needs ? You bet !! Especially since this cpu was VERY easily overclockable from 2.13Ghz stock to 3.2GHZ on factory aircooling. THAT was the real payoff in terms of value. Would I buy a quad core machine today ? Well, it depends on the dollars (value) entirely. If I could fold with 4 cores at the same rate as my Allendale for a lower cost per core, it is a done deal !

RPhArrow --\\\-------------------------->

 

[FrgMstr]

Typo fixed.

The Xbox 360 has three dual-threaded processors onboard,

 

[enumae]

Please take the time and read, it took a little while to write

These numbers are from http://www.tomshardware.com/2006/09/10/four_cores_on_the_rampage/page13.html a review of the 2.6GHz Kentsfield on a 1066 FSB.

They are just a base for the example of future performance of K8L at 2.6GHz. I understnad there are alot of other factors and if you know what they are please explain, especially you Sharikou.

Remember these numbers are just a base, hence not real.

Some have claimed...

"On performance of the cores, we already know, 60% increase in integer speed, and 200% increase on floating point."

X6800 at 2.93GHz SiSoft 2007 CPU test Multimedia floating point.

87,615/2 = 43,808 FP per core ---- 2.8*100/2.93=0.95*43,808=41,618 for a 2.8GHz Conroe Core

41,618*4 = 2.8GHz Kentsfield Quad core 166,472 vs 2.6GHz Kentsfield Quad core 158,733 as per benchmark.

FX62 at 2.8GHz SiSoft 2007 CPU test Multimedia floating point.

57,526/2 = 28,763 fp per core * 2 (for the 200% increase in FP) = K8L per core 57,526*4 = K8L Quad core 230,104

<b>The AMD K8L at 2.8GHz has a 28% performance lead in Multimedia Floating Point.</b>

X6800 at 2.93GHz SiSoft 2007 cpu test Multimedia Integer.

162,469/2 = 81,235 integer per core --- 2.8*100/2.93=0.95*81,235=77,173 for a 2.8GHz Conroe core

77,173*4 = Kentsfield Quad core 308,691 vs 2.6GHz Kentsfield Quad core 294,383 as per benchmark.

FX62 at 2.8GHz SiSoft 2007 cpu test Multimedia Integer

52,869/2 = 26,434 integer per core * 1.6 (for the 60% increase in Integer performance) = K8L per core 42,295*4 = K8L Quad core 169,180

<b>The Intel Kentsfield Quad core at 2.8GHz has a 45% performance lead in Multimedia Integer.</b>


Now for the clock speeds, some have said 4GHz for K8L, please bare with me...

Assuming an AMD 3.6GHz CPU is at about 150W TDP (2.8*100/3.6=1.23% increase so 2.8*1.23= 150W +/-), so at 4.0GHz its about a TDP of 160W at 90nm (straight scale).

I am keeping this really simple, so the numbers may be off but they give a decent idea...

65nm*100/90nm = 0.72

160W*0.72 = 115W 4.0GHz at 65nm.

While it may be possible for a K8L dual core, it is not likely for K8L Quad core.

For K8L to run at 4.0GHZ would have a TDP of roughly 115w*1.75 (assuming it would not just double) = 200W.

125w*100/200w=0.62

4.0GHz*0.62=2.5GHz

Theoretcally a 2.5 - 2.6GHz K8L is possible in the 125 - 130W envelope that is currently used by AMD.

Keep in mind we have not even factored in the 4MB of L2 and L3 cache or the improved manufacturing process AMD plans to use so yes the numbers could get better or worse.

Now what kind of performance advantage does K8L have at 2.6GHz...

2.6*100/2.8=0.93

<b>Multimedia floating point.</b>

2.6GHz Kentsfield Quad core 158,733 vs 2.6GHz K8L Quad core 230,104*0.93 = 213,997

The 2.6GHz K8L Quad core has a 26% advantage over the 2.6GHz Kentsfield Quad core.

<b>Multimedia Integer</b>

2.6GHz Kentsfield Quad core 294,383 vs 2.6GHz K8L Quad core 169,180*0.93 = 157,337

The 2.6GHz Kentsfield has a 46% advantage over the 2.6GHz K8L Quad core.

The battle does not seem to be over, but just beginning.

Next year should be an amazing time for any of us who enjoy technology not just AMD or Intel.

 

[Unknown-One]

Please take the time and read, it took a little while to write

These numbers are from http://www.tomshardware.com/2006/09/10/four_cores_on_the_rampage/page13.html a review of the 2.6GHz Kentsfield on a 1066 FSB.

They are just a base for the example of future performance of K8L at 2.6GHz. I understnad there are alot of other factors and if you know what they are please explain, especially you Sharikou.

Remember these numbers are just a base, hence not real.

Some have claimed...

"On performance of the cores, we already know, 60% increase in integer speed, and 200% increase on floating point."

X6800 at 2.93GHz SiSoft 2007 CPU test Multimedia floating point.

87,615/2 = 43,808 FP per core ---- 2.8*100/2.93=0.95*43,808=41,618 for a 2.8GHz Conroe Core

41,618*4 = 2.8GHz Kentsfield Quad core 166,472 vs 2.6GHz Kentsfield Quad core 158,733 as per benchmark.

FX62 at 2.8GHz SiSoft 2007 CPU test Multimedia floating point.

57,526/2 = 28,763 fp per core * 2 (for the 200% increase in FP) = K8L per core 57,526*4 = K8L Quad core 230,104

The AMD K8L at 2.8GHz has a 28% performance lead in Multimedia Floating Point.

X6800 at 2.93GHz SiSoft 2007 cpu test Multimedia Integer.

162,469/2 = 81,235 integer per core --- 2.8*100/2.93=0.95*81,235=77,173 for a 2.8GHz Conroe core

77,173*4 = Kentsfield Quad core 308,691 vs 2.6GHz Kentsfield Quad core 294,383 as per benchmark.

FX62 at 2.8GHz SiSoft 2007 cpu test Multimedia Integer

52,869/2 = 26,434 integer per core * 1.6 (for the 60% increase in Integer performance) = K8L per core 42,295*4 = K8L Quad core 169,180

The Intel Kentsfield Quad core at 2.8GHz has a 45% performance lead in Multimedia Integer.


Now for the clock speeds, some have said 4GHz for K8L, please bare with me...

Assuming an AMD 3.6GHz CPU is at about 150W TDP (2.8*100/3.6=1.23% increase so 2.8*1.23= 150W +/-), so at 4.0GHz its about a TDP of 160W at 90nm (straight scale).

I am keeping this really simple, so the numbers may be off but they give a decent idea...

65nm*100/90nm = 0.72

160W*0.72 = 115W 4.0GHz at 65nm.

While it may be possible for a K8L dual core, it is not likely for K8L Quad core.

For K8L to run at 4.0GHZ would have a TDP of roughly 115w*1.75 (assuming it would not just double) = 200W.

125w*100/200w=0.62

4.0GHz*0.62=2.5GHz

Theoretcally a 2.5 - 2.6GHz K8L is possible in the 125 - 130W envelope that is currently used by AMD.

Keep in mind we have not even factored in the 4MB of L2 and L3 cache or the improved manufacturing process AMD plans to use so yes the numbers could get better or worse.

Now what kind of performance advantage does K8L have at 2.6GHz...

2.6*100/2.8=0.93

Multimedia floating point.

2.6GHz Kentsfield Quad core 158,733 vs 2.6GHz K8L Quad core 230,104*0.93 = 213,997

The 2.6GHz K8L Quad core has a 26% advantage over the 2.6GHz Kentsfield Quad core.

Multimedia Integer

2.6GHz Kentsfield Quad core 294,383 vs 2.6GHz K8L Quad core 169,180*0.93 = 157,337

The 2.6GHz Kentsfield has a 46% advantage over the 2.6GHz K8L Quad core.

The battle does not seem to be over, but just beginning.

Next year should be an amazing time for any of us who enjoy technology not just AMD or Intel.

Fixed bold tags, made the text size 4, and turned the headings red for extra visibility

Note that this forum uses brackets for tags, like [b ] and [/b ] instead of <b> and </b>

 

[Amiteriver]

Have a 4X4 question didn't want to start a new thread
I presently have a AM2 4000x2 Was wondering if anybody knows if 2 of these would work in the 4x4 or will you have to use some of the new FX combos that I read about somewhere.

 

[Mohonri]

Those numbers are very interesting--it seems that AMD has held the FP performance advantage for quite a while, while Intel has held the Integer performance crown. And that seems to be continuing.

The other interesting thing is that Intel quad-core will be coming out before AMD quad-core. Does that mean that during that gap, Intel will ramp up clock speeds enough to eliminate the AMD advantage in FP?

 

[drizzt81]

Have a 4X4 question didn't want to start a new thread
I presently have a AM2 4000x2 Was wondering if anybody knows if 2 of these would work in the 4x4 or will you have to use some of the new FX combos that I read about somewhere.

It would appear that new(er) chips are required for 4x4, since most desktop chips have an insufficient number of hypertransport links.

 

[shaun]

Those numbers are very interesting--it seems that AMD has held the FP performance advantage for quite a while, while Intel has held the Integer performance crown. And that seems to be continuing.

The other interesting thing is that Intel quad-core will be coming out before AMD quad-core. Does that mean that during that gap, Intel will ramp up clock speeds enough to eliminate the AMD advantage in FP?

It's possible - however granted as Kyle stated, the Intel quad-core will be 2 conroes, while the AMD will be a native quad-core.

As for performance, I don't think simple ramping of clock speeds will gain Intel too much ground without some sort of architectural upgrade as well.

We all saw what happened when Intel released their "dual core" processors to compete with AMD's first dual-core.

Speed didn't save them there, Conroe did.

 

[mpcamer1220]

Have a 4X4 question didn't want to start a new thread
I presently have a AM2 4000x2 Was wondering if anybody knows if 2 of these would work in the 4x4 or will you have to use some of the new FX combos that I read about somewhere.

You will not be able to, for a couple of reasons..

One, you have the wrong socket - 4X4 won't use AM2.

Secondly, your chip doesn't have enough HyperTransport links.

 

[duby229]

You will not be able to, for a couple of reasons..

One, you have the wrong socket - 4X4 won't use AM2.

Secondly, your chip doesn't have enough HyperTransport links.

Everything I've seen so far suggested it will be dual socket AM2 with NUMA. That they would not be sold as individual chips, but instead as a matched set that will be branded as FX.

 

[mpcamer1220]

Everything I've seen so far suggested it will be dual socket AM2 with NUMA. That they would not be sold as individual chips, but instead as a matched set that will be branded as FX.

Check this post out - http://hardforum.com/showpost.php?p=1029988862&postcount=22


The guy I was on the phone with was Damon Muzny.. He's pretty high up in the PR Department at AMD..

He didn't flat out say that 4X4 would not be Socket AM2, but he did imply that it might not be.. He said that a lot of people are assuming that it will be Socket AM2, and that they think they can go buy one processor now, and use it once 4X4 is available.. He then told me that people should not do that..

 

[duby229]

Check this post out - http://hardforum.com/showpost.php?p=1029988862&postcount=22


The guy I was on the phone with was Damon Muzny.. He's pretty high up in the PR Department at AMD..

He didn't flat out say that 4X4 would not be Socket AM2, but he did imply that it might not be.. He said that a lot of people are assuming that it will be Socket AM2, and that they think they can go buy one processor now, and use it once 4X4 is available.. He then told me that people should not do that..

Check this out if you dont believe me. I know what he said, it prolly has something to do with socket F.


http://tweakers.net/nieuws/44504/AM...erd-op-Socket-1207.html?www.reghardware.co.uk

 

[MrWizard6600]

i dont mean to ruffle to many feathers...

but it seems to me, for the people who do need 4 funcional processers, amds solution is better. i mean i thought intel learned that perhalps the cores on seperate dies thing wasnt such a good idea ever since the D 600's. apparently not.

but who knows, four cores in any arrangement is still four cores, i just wanna see Barcelona (nice name... first time ive heard it) go head 2 head with kentsfeild, and Grayhound go Head 2 head with Conroe and Allendale. should be one helluva'n interesting fight.

but thats another thing that accured to me, die processing. amds best is 65nm right now, intels best, 45, Nvs best (yes cpu is not a gpu but they both use a simliar wafer/die system) is 90 and ATis best is 90, the engineers that have been working on reducing die size has been, well effectivly doubled. once the merger is complete, i think were gonna start to see amd/ati pull ahead in the shrinking race --weather that means a serios improovement in performance, i dont know.

and imho, ati has the bright side too. the best rumers iv heard have all mentioned the G80 consisting of a rather effed up architecture, and using upward of 700mil transistors...
"wait a minute, thats good right? more transistors is more processes and more pipes right?"
sure but it also costs more, the G71 managed the same number of processes as a G70 with 40mil less transistors (310: G70, 270: G71). this is one of the major reasons the G71 could release at such a nice price. (i mean really, the 7800gtx is a dead card, you can get a higher clocked, better performing, cooler running, cheaper card, the 7900GT.

as for the R600s architecture, well the rumers differ to greatly for me to make a really educated guess... but i havnt heard of such an obcene ammount of transistors.

you should have asked them about the R600!!

 

[enumae]

but it seems to me, for the people who do need 4 funcional processers, amds solution is better. i mean i thought intel learned that perhalps the cores on seperate dies thing wasnt such a good idea ever since the D 600's. apparently not.

I believe that in the past the reason that Intel failed with the Pentium D, was not the approach, but the cores themselves.

Conroe is a completely different core and looking at the benchmarks, seem to be scaling at about 65-75%, thats not bad.

AMD might see 80%, but K8L is also not going to be competing with Kentsfield or Cloverton, K8L is going up against Nehalem (server) and Bloomfield (desktop and mobile).

There are a few articles/blogs which have a very good explanation why AMD can not or will not just glue two cores together, but here is one.

Why AMD has to go native.

 

[Lolento]

i dont mean to ruffle to many feathers...

but it seems to me, for the people who do need 4 funcional processers, amds solution is better. i mean i thought intel learned that perhalps the cores on seperate dies thing wasnt such a good idea ever since the D 600's. apparently not.

but who knows, four cores in any arrangement is still four cores, i just wanna see Barcelona (nice name... first time ive heard it) go head 2 head with kentsfeild, and Grayhound go Head 2 head with Conroe and Allendale. should be one helluva'n interesting fight.

but thats another thing that accured to me, die processing. amds best is 65nm right now, intels best, 45, Nvs best (yes cpu is not a gpu but they both use a simliar wafer/die system) is 90 and ATis best is 90, the engineers that have been working on reducing die size has been, well effectivly doubled. once the merger is complete, i think were gonna start to see amd/ati pull ahead in the shrinking race --weather that means a serios improovement in performance, i dont know.

and imho, ati has the bright side too. the best rumers iv heard have all mentioned the G80 consisting of a rather effed up architecture, and using upward of 700mil transistors...
"wait a minute, thats good right? more transistors is more processes and more pipes right?"
sure but it also costs more, the G71 managed the same number of processes as a G70 with 40mil less transistors (310: G70, 270: G71). this is one of the major reasons the G71 could release at such a nice price. (i mean really, the 7800gtx is a dead card, you can get a higher clocked, better performing, cooler running, cheaper card, the 7900GT.

as for the R600s architecture, well the rumers differ to greatly for me to make a really educated guess... but i havnt heard of such an obcene ammount of transistors.

you should have asked them about the R600!!

this is off topic but AMD and Intel has internal fab development while Nvidia and ATi rely largely on TSMC...

there is a lot of politics going on here as places like TSMC and UMC sell out their capacity 3-months to 6-months ahead of time

if ATI intends to go internal with AMD, AMD has better be able to swallow all of ATi's demand because fab subcons are capacity contrained most of time and it would make more business-sense to sell to someone who has repeat business (ie Nvidia) than to sell to someone who is only off-loading capacity

 

[drizzt81]

Why AMD has to go native.

I wish that link was in English.

 

[Martyr]

You are right, why in the heck would anyone want a system where they could add a second GPU for more graphics capabilities, or another CPU for more SMP performance enhancements.

Monolithic FTW!

While it may not be for everyone, I personally have the exact opposite view.

1. While I'm not using SLI, I purchased an SLI motherboard so that if I ever wanted to, I could pickup a second card and have enhanced performance. This covers me for in between GPU generations as the early adopters will sell their older cards on ebay for fairly cheap. I've seen lots of arguments from non-SLI or CrossFire adopters that it's insane to buy two older gen cards when a single new-Gen card may provide similar performance. YET, this market segment is growing so they must all be idiots like myself

2. Upgradeability of CPU performance by adding another processor. This personally excites me as instead of having to wait for an Octa-Core single processor, I can add two of the quad-core (or add two dual-core for quad-core like performance). This makes even more sense than SLI or CrossFire in my opinion. I could for example, purchase a 4x4 motherboard and one single CPU of the best performance that I could afford at that time. Another year or so down the road, I could throw in a second at a much cheaper price (especially since AMD and Intel will be locked into pricing wars to win the CPU battle).

Having those options available doesn't mean you are forced to exercise them, but it provides additional options to the consumer, which is always a win.

If like SLI it catches on, and I'll personally be one of the purchasers of a 4x4 motherboard, you can just sit back and tell us all how lame we are when we could purchase the next gen single proc system for less (and have to purchase them more frequently to keep a reasonable performance crown).

the problem with sli type systems is price. they are only effective if the prices drop fast enough on the old components to make buying an older card for a 50% performance gain is better than buying a new card for possibly a much larger performance gain. the prices on older cards do not drop for shit. and the same thing applies to many older processors.

 

[enumae]

I wish that link was in English.

It seems to be if I click on it.

 

[Martyr]

It seems to be if I click on it.

try reading it.

 

[drizzt81]

While a MCM may be smaller and may be more "elegent" in some respects,

that cracked me up, lol!

try reading it.

that was my point exactly

if ATI intends to go internal with AMD, AMD has better be able to swallow all of ATi's demand because fab subcons are capacity contrained most of time and it would make more business-sense to sell to someone who has repeat business (ie Nvidia) than to sell to someone who is only off-loading capacity

Given the time-value of money, a deal now is often worth a lot more than a deal inthe future. I am sure that neither UMC nor TMSC would turn away a paying customer. They are not in the market to favor "green over red" or something, but to provide a service to customers. Let's also not forget that the negotiations between their customers and the fabs are unlikely to be single-issue negotiations, which means that the potential contract area is likely large...

 

[Lolento]

Given the time-value of money, a deal now is often worth a lot more than a deal inthe future. I am sure that neither UMC nor TMSC would turn away a paying customer. They are not in the market to favor "green over red" or something, but to provide a service to customers. Let's also not forget that the negotiations between their customers and the fabs are unlikely to be single-issue negotiations, which means that the potential contract area is likely large...

Yes, of course, if there is money why not take it?

But what I'm saying is that when the business is good (semicon up-turn which is around a 5-year cycle), these subcons are capacity contrained. Capacity priority is given to fab-less design houses that are fully dependent on the subcons.

That's why right after the announcement of the AMD purchase, ATi had to jump out and say they are still partners with TSMC and UMC. If they didn't, capacity at those places will be allocated to someone else. The subcons sell out their capacities months ahead of time. It's not like if you want to get X number of wafers fabbed in 12 weeks, you can send out a purchase order today; it doesn't work like that.

 

[GoodOmens]

The other interesting thing is that Intel quad-core will be coming out before AMD quad-core. Does that mean that during that gap, Intel will ramp up clock speeds enough to eliminate the AMD advantage in FP?

Intel already has plans for a native quad core chip about a quarter after AMD releases their's.

 

[drizzt81]

But what I'm saying is that when the business is good (semicon up-turn which is around a 5-year cycle), these subcons are capacity contrained. Capacity priority is given to fab-less design houses that are fully dependent on the subcons.

I understood your point. I am sure that negotiations between the fabricators and their respective customers will include that. It is likely that ATi may have to pay a "uncertain future" premium if it requires fab capacity, while nVidia can agree to a more long-term contract. I am sure that once this time comes, we will see or hear about the results, even if indirectly.

 

[Thin Man]

I thought 4x4 would either be two Socket Fs on a mobo or two Socket AM2s on a mobo. I am under the impression that a native quad-core will be available on Socket AM2 in the next year or so. Correct?

 

[Martyr]

I thought 4x4 would either be two Socket Fs on a mobo or two Socket AM2s on a mobo. I am under the impression that a native quad-core will be available on Socket AM2 in the next year or so. Correct?

yes, about 6 months after intel releases a double duel core chip, and also about the same time intel will release their native quad core solution.

 

[drizzt81]

I thought 4x4 would either be two Socket Fs on a mobo or two Socket AM2s on a mobo. I am under the impression that a native quad-core will be available on Socket AM2 in the next year or so. Correct?

From all the information that I have seen you are correct.

 

[(cf)Eclipse]

i'm surprised nobody linked this...
http://www.theinquirer.net/default.aspx?article=34742

 

[J-M-E]

So if I bought an AM2 motherboard with an am2 dual core processor now, I wouldnt have to change out my mobo when quad core comes out?

 

[mpcamer1220]

So if I bought an AM2 motherboard with an am2 dual core processor now, I wouldnt have to change out my mobo when quad core comes out?

Well, with Socket 939; you could just pop in an X2.. so I'm assuming that AM2 will be the same with with quad core.. Henri Richard said "and of course, AM2 will support quad core.." or something along those lines, awhile back..