More on AMD's roadmap

[Viditor]

Inquirer

X2 to 3GHz in November @90nm,
X2 to 2.6GHz in December @65nm (though I would bet that these OC better),
Sempron 3500 (2GHz@35w) and 3800 (2.2GHz@62w) @90nm in October

 

[dexvx]

Is the X2 @ 65nm debuting at a lower clock or is that just the LOWEST clock its going to debut at?

 

[HopJokey]

Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

 

[Viditor]

I think that a lot of the reason for the lower clocks on the initial X2s have to do with the way AMD's manufacturing works (see this thread).
Some other possibilities include:
1. limited 65nm production may mean that they will be releasing another line at a low cost specifically for 4x4
2. They are going to reduce prices even more on X2s for the 65nm...this makes the most sense to me as even with higher clocks, they won't match Conroe in the high end until Rev H is released. By keeping a similar architecture and using the 65nm process with a lower bin speed, they are maximizing their profit on the chips (higher bins produce lower yields) which could allow them to really drop their prices on X2 again and still make a very nice profit.

 

[imported_Questar]

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

 

[FelixDeCat]

Is anybody taking preorders for X2 to 2.6GHz in December @65nm? Does anybody want to preorder?

 

[Viditor]

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

It may have been discussed by some, but the yield problem rumour has been shown to be nothing but FUD...
It was first reported by Charlie D, and retracted within 48 hours...

 

[BrownTown]

Nobody but a few select people at AMD really know exactly what the yield numbers are, so anything else is jsut speculation by more or less informed people, but still speculation. It is not uncommon for a new process to debut at slower clockspeeds than the old process. This allows them to get good yields of lower end parts till the process is tweaked enough to get it working just right. The fact that the 65nm parts are 400mhz slower means that AMDs 65nm process is not doing as well as some might hope, but says little about wherre 65nm clockspeeds will be in 6 months time. Don't forget that Intels 65nm parts also came in at lower clockspeeds than their 90nm parts, but they quickly surpassed the 90nm parts.

 

[Hard Ball]

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

Actually, the norm for going to a new node for CPU dice is to clock them lower than the contemporary dice from the old process. Think back for the first Prescotts in Feb 04, which were only 2.8GHz for volume shipping; and for the first 90nm Winchesters in Sept 04, where 2.2GHz vs the 2.6 GHz FX55 on 130nm and 2.4GHz 3800+ also on 130nm.

 

[dexvx]

Originally posted by: Hard Ball
Actually, the norm for going to a new node for CPU dice is to clock them lower than the contemporary dice from the old process. Think back for the first Prescotts in Feb 04, which were only 2.8GHz for volume shipping; and for the first 90nm Winchesters in Sept 04, where 2.2GHz vs the 2.6 GHz FX55 on 130nm and 2.4GHz 3800+ also on 130nm.

Not necessarily.

Willamettes@180nm were clocked lower than Northwoods@130nm.
P3 Katmai's@350nm were clocked lower than the 250nm Coppermine
Dothans@90nm were clocked higher than 130nm Banias

When most people think die shrink (with no arch changes), they generally think of faster clock speed.

 

[The Mailman]

im not upgrading until 4ghz @65

 

[imported_Questar]

Originally posted by: Viditor

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

It may have been discussed by some, but the yield problem rumour has been shown to be nothing but FUD...
It was first reported by Charlie D, and retracted within 48 hours...

Nope.

Look at my post on the subject. I think it was July when my financial advisor sent me the research report, a full two months ahead of Charlie's article.

If you would like to verify, I'm sure you have access to these types of reports. I think mine was $85.

 

[Viditor]

Originally posted by: Questar

Originally posted by: Viditor

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

It may have been discussed by some, but the yield problem rumour has been shown to be nothing but FUD...
It was first reported by Charlie D, and retracted within 48 hours...

Nope.

Look at my post on the subject. I think it was July when my financial advisor sent me the research report, a full two months ahead of Charlie's article.

If you would like to verify, I'm sure you have access to these types of reports. I think mine was $85.

Which report? I do indeed have access to many reports (including IDC and Mercury which costs me ~$20k/year). But not one report has even mentioned this in a serious manner...

 

[Viditor]

Originally posted by: Questar

Originally posted by: Viditor

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

It may have been discussed by some, but the yield problem rumour has been shown to be nothing but FUD...
It was first reported by Charlie D, and retracted within 48 hours...

Nope.

Look at my post on the subject. I think it was July when my financial advisor sent me the research report, a full two months ahead of Charlie's article.

If you would like to verify, I'm sure you have access to these types of reports. I think mine was $85.

Which report? I do indeed have access to many reports (including IDC and Mercury which costs me ~$20k/year). But not one report has even mentioned this in a serious manner...

 

[OcHungry]

So what if it's been reported before? why is that so important? and why so much fuss about 65nm yeild or only 2.6ghz WTF?
get over it. AMD is on schedule and before you know it K8L's will be released and kick ass.
Time for me to start thinking about upgrading to AM2 platform

 

[StrangerGuy]

So I guess AMD will still get their ass kicked in the performance sector by Intel this year...130nm to 90nm only gave K8 an additional 400MHz headroom. I bet the 65nm will give at most another 400MHz extra which will give AMD the edge to barely outrun the E6700 with a 3.4GHz A64, that is provided that they got their 65nm right.

 

[harpoon84]

Originally posted by: StrangerGuy
So I guess AMD will still get their ass kicked in the performance sector by Intel this year...130nm to 90nm only gave K8 an additional 400MHz headroom. I bet the 65nm will give at most another 400MHz extra which will give AMD the edge to barely outrun the E6700 with a 3.4GHz A64, that is provided that they got their 65nm right.

Was that ever in doubt? 65nm K8s were never going to defeat C2D.

4x4 however might come closer to Kentsfield levels than many people imagine.

2 x 3GHz FXs in 4x4 might come pretty close to a 2.66GHz Kentsfield, assuming 4x4 exhibits better scaling than the FSB constrained Kentsfield.

However, 2 x 125W TDP FXs is asking for trouble on the efficiency/heat side of things.

 

[StrangerGuy]

Originally posted by: harpoon84

Originally posted by: StrangerGuy
So I guess AMD will still get their ass kicked in the performance sector by Intel this year...130nm to 90nm only gave K8 an additional 400MHz headroom. I bet the 65nm will give at most another 400MHz extra which will give AMD the edge to barely outrun the E6700 with a 3.4GHz A64, that is provided that they got their 65nm right.

Was that ever in doubt? 65nm K8s were never going to defeat C2D.

4x4 however might come closer to Kentsfield levels than many people imagine.

2 x 3GHz FXs in 4x4 might come pretty close to a 2.66GHz Kentsfield, assuming 4x4 exhibits better scaling than the FSB constrained Kentsfield.

However, 2 x 125W TDP FXs is asking for trouble on the efficiency/heat side of things.

Contrary to most people would believe, Kentsfield is never FSB starved even with a 1066MHz FSB. Proof

Besides, Intel can easily up the FSB to 1.6GHz or faster in the future(400MHz base) if Kentsfield needs more FSB bandwidth since current 965P chipsets and E6300/E6400 can handle those speeds.

 

[harpoon84]

Yes but I'm sure 4x4 will still scale slightly better than Kentsfield.

Yorkfield will be 1.33GHz FSB, better, but it's still no IMC.

 

[Furen]

Originally posted by: StrangerGuy
Contrary to most people would believe, Kentsfield is never FSB starved even with a 1066MHz FSB. Proof

Besides, Intel can easily up the FSB to 1.6GHz or faster in the future(400MHz base) if Kentsfield needs more FSB bandwidth since current 965P chipsets and E6300/E6400 can handle those speeds.

Huh? Intel can increase the FSB to 400MHz with two loads on it? I wonder... I dont doubt that Intel MAY be able to get 400MHz out of its chipsets with Conroes but having two chips on the same FSB does decrease clock potential.

 

[hans007]

its obvious the yield is not doing that great. intel's 90nm transition was pretty bad also, but that was because that was just bad in general not to mention the prescott was not a die shrink it was a completely different cpu with 10 more pipeline stage.

youc ant compare that really.

the x2 65nm is a die shrink. it SHOULD be doing at least the 90nm speeds. the problem is this is the same proces that ibm/motorola are having problems with as they codeveloped it and amd uses the same one. Cell is also having problems and the xbox 360 cpu is also not at 65nm yet and they are all ibm fabbed on similar process.

intel went and did it on its own. the 65nm cpus were 3.6ghz from the start actually the intel 960, though there were 3.8 ghz 90nm ones. not really a fair comparison though since cedar mill didn't really ship that many units for single core, and most of the high end 65nm cpus were dual core dual die units which made more heat (thus the lower clock speed). they also had more cache.

 

[Furen]

Originally posted by: hans007
its obvious the yield is not doing that great. intel's 90nm transition was pretty bad also, but that was because that was just bad in general not to mention the prescott was not a die shrink it was a completely different cpu with 10 more pipeline stage.

youc ant compare that really.

the x2 65nm is a die shrink. it SHOULD be doing at least the 90nm speeds. the problem is this is the same proces that ibm/motorola are having problems with as they codeveloped it and amd uses the same one. Cell is also having problems and the xbox 360 cpu is also not at 65nm yet and they are all ibm fabbed on similar process.

intel went and did it on its own. the 65nm cpus were 3.6ghz from the start actually the intel 960, though there were 3.8 ghz 90nm ones. not really a fair comparison though since cedar mill didn't really ship that many units for single core, and most of the high end 65nm cpus were dual core dual die units which made more heat (thus the lower clock speed). they also had more cache.

Hardly "obvious"... AMD did not have any serious problems with 90nm and it still started with lower-end parts first, then moving onto higher-end parts. It makes sense to start with higher-volume parts because it allows AMD to increase margins on these parts, and perhaps introduce cheaper parts, too. Of course, it also allows AMD to not worry about any potential problems that may affect yields at higher binnings, the CPU market doesn't like Phantom Edition products like the video market seems to. Then there's also the fact that AMD said quite a few times that it wanted to have 65W TDPs for the 65nm parts, which may hold back clock speeds, at least for a while. Intel's 90nm problems were no such thing. Even initial Prescotts could hit 4GHz+ without too much trouble. The problem was that the architecture produced heat like mad, so Intel was held back not by the architecture itself but by the heat it produced. Right now Intel is also being held back by heat, but not in the same way. Intel is sticking to its TDPs strictly, which limits its ability to push clocks higher. While all kinds of Conroes are hitting 3.5GHz+ clock speeds, I seriously doubt they can achieve that clock with less than a 100W power draw (I cant say for certain since I dont have one yet, nor do I plan to measure power draw when I do get one).

 

[imported_Questar]

AMD did not have any serious problems with 90nm

Go back and read the news. AMD was year late with 90nm.

 

[imported_Questar]

Originally posted by: Viditor

Originally posted by: Questar

Originally posted by: Viditor

Originally posted by: Questar

Originally posted by: HopJokey
Good find Viditor.

One question I have is why aren't the 65nm parts clocked higher? Will they once the 65nm ramp is further along?

It's been discussed before, AMD is having yeild problems at 65nm, and has been for months.

It may have been discussed by some, but the yield problem rumour has been shown to be nothing but FUD...
It was first reported by Charlie D, and retracted within 48 hours...

Nope.

Look at my post on the subject. I think it was July when my financial advisor sent me the research report, a full two months ahead of Charlie's article.

If you would like to verify, I'm sure you have access to these types of reports. I think mine was $85.

Which report? I do indeed have access to many reports (including IDC and Mercury which costs me ~$20k/year). But not one report has even mentioned this in a serious manner...

Well there's your problem. You're getting your reports from the IT analysts and not the financial analysts. I use Gartner for IT product and process reports. I use a full sevice investment banker for financial reports.

 

[oRdchaos]

I would imagine they're starting at lower clock speeds so that they can ship in volume at the frequencies they're stating immediately.

If they're going to release at 3ghz at 90nm, they've been binning processors that run at that speed for a while.