AMD 65nm chips will arrive in H1 2007

[BrownTown]

well, obviously big red flags should come up when Intel says that its 45nm has only 20% better transister performance OR 1/5 the leakage of the current process (since trigate + high-k should provide much more then that). But an Intel proceess engineer confirmed that all this is not in the current 45nm process. So its pretty much the same as the 65nm except smaller. Still uses dry lithograhy etc...

so i'd expect that leakage will be a very high percent of total power consuption since as the process gets smaller leakage goes up and dynamic power goes down. Most likely no better then current 90nm in terms of heat...

 

[dmens]

Who's the process engineer? Giving away details on intel's most highly guarded secret... tsk tsk. Now, consider the fact that device count and total Z on a 45nm product would be far higher than a 65nm or 90nm part, a 1/5 total reduction in leakage is expected with the technology used. As for device delays, please explain why 45nm "should" be "much more" than 20% faster than 65nm.

 

[BrownTown]

well, for one thing its not a secret, its been published by Intel, but obviously they aren't goign around bragging about having to cut their new techs. As for the leakage, thats 1/5 the leakage, not 1/5 reduction, in other words, 20% of current, not 80%, but you only get that when you sacrifice speed for power. Also, i'm not the one saying ti should be mroe then 20% faster, that was Intels hype about trigate which will now nto be included. Furthermore, it should be noted that 20% better transister performance is completely differnet the 20% mroe clockspeed.

 

[dmens]

Sorry, I meant 1/5 in leakage compared to before. Is that only with a high Vt? I doubt it.

 

[stevty2889]

Originally posted by: BrownTown
yeah, process is EVERYTHING, the architecture only becomes important when theprocess technology is close to the same, presscott didn't suck becasue of its architecture (although that was certainly part of it), the problem with presscott was the crappy 90nm tech leaked so much that they couldn't scale it up nearly as high as the aritecure requried in order to be efficient.

Dothan is made on the same 90nm proccess..and doesn't have the problems of Prescott..Prescotts problem was not the proccess..

 

[BrownTown]

dothan runs at a lower clockspeed and voltage? then presscott, so it produces alot less power. In order for dothan to be competative it only has to be around 2.0G, in order for presscott to compete with desktop processors it has to work at 4+G, the power consumed by the 90nm at 4G is too much for normal air cooling to effectively deal with. If presscott were able to run at 4.5G without producing so much power then it wouldn't suck...

 

[Viditor]

Originally posted by: dmens
Make things up? No, everything I said is based on knowledge of the industry from working in cpu design. I might work for intel, but I'm not going to lie to myself or anyone about the straight industry facts. Process is king because it plays to the economies of scale.

Don't start with the tired old story that Intel's marketing power is the sole reason it keeps market share. If AMD were able to produce enough volume to supply the big boys, they'd get all the brand recognition in the world simply from exposure. The fact that they cannot compete at that level of volume means that they will have zero brand recognition to the average consumer, and no lawsuit will change that. Public perception? Most of the public hasn't even heard of AMD. Why do you think AMD is going through the lawsuit to begin with?

Then I assume that with all of AMD's extra volume coming on-line next quarter (Fab36 shipping for volume on 300mm, and Charter shipping AMD for revenue on 300mm), you feel that AMD will probably start selling to the only "big boy" OEM they aren't already selling to?

 

[BrownTown]

meh, any how many 300mm plants does Intel have? a whole lot methinks, several at 65nm which AMD has none, and 2 headed for 45nm which also AMD has none, but anyways, they still, if they keep having better processors they will get the $$$ to build more fabs and produce some serious volume. Plus, all they produce is processors, while Intels fabs are producing all kinda of other stuff too like chipsets and integrated gfx, and all of Intels other stuff

 

[Viditor]

meh, any how many 300mm plants does Intel have?

True, but that's not really the point...AMD has jumped to ~21% marketshare with their single 200mm plant. With 2 more 300mm Fabs on-line next quarter, that gives them capacity for at least a 50% marketshare this year. I really don't think AMD will be capacity limited ever again (which is all that really matters).


BTW, to go back to the original point of the thread, Phil Hester (AMD's CTO) made public comments today on 65nm...

What I said last November at our annual analyst meeting still holds. We have had 65nm preliminary silicon running in our Dresden Fab 36 since last June; we plan to begin 65nm volume production in the second half of 2006, and we plan to be substantially converted to 65nm in Fab36 by mid-2007

Hester Interview

 

[BrownTown]

k, and Intel had 65nm volume production in 05, so about a year behind (probably less). But yeah, AMD maybe have the capacity to get to 50% marketshare (i don't know, but lets assume thats true), but still they will not reach that point because I still feal that in many people eyes (though the number gets smaller everyday), they are a second rate product. I know, for example, I did an presentation on AMD vs Intel for a technical writing/speaking class last semester, and the professor (a chemical engineer), was still pretty convinced that AMD processors werent good for anything but games, and still bought into the clockspeed myth. So, if engineering professors and major universites don't all know AMDs advantage, then AMD still has a long way to go in convincing people.

Also, AMDs performance lead should shrink some this year, so it will be harder to convince peopel to switch if Intel CPUs are actually splitting benchmarks 50-50 with AMD instead of losing 70% of single core and 90% of dual core benchies.

 

[Viditor]

Intel had 65nm volume production in 05, so about a year behind (probably less)

About 6 months if you measure by release dates...but you have to keep in mind a few things:
1. AMD has no need to release 65nm, while Intel definately does
2. 65nm has 2 major benefits, cost/chip and (in this case) power reduction. However, most people forget that in the case of cost, yields are an even bigger factor. AMD is currently enjoying over 85% yield on their 90nm process (an exceptionally high yield). By delaying their volume production of 65nm while simultaneously improving their yields on it, AMD is able to optimize their return for the process. In fact, they have a superb bit of software (called APM) that calculates the best time for the transition.
3. AMD did exactly the same thing with 90nm...they began volume production at close to mature yields. They have stated that 65nm volume production will START at already mature yields...

still they will not reach that point because I still feal that in many people eyes (though the number gets smaller everyday), they are a second rate product.

I agree that this is the biggest hurdle (perception). The only upside is that AMD is now being pushed by every OEM except Dell (and who knows when/if they will convert). The key indicator here is that AMD has gone from 6 Fortune 100 companies as clients to over 90 in the last 2 years. Consumers are a very small part of the equation...

 

[openwheelformula1]

"The key indicator here is that AMD has gone from 6 Fortune 100 companies as clients to over 90 in the last 2 years."

I was not aware of this great achievement. Wow.

 

[RaynorWolfcastle]

Originally posted by: Viditor

Intel had 65nm volume production in 05, so about a year behind (probably less)

About 6 months if you measure by release dates...but you have to keep in mind a few things:
1. AMD has no need to release 65nm, while Intel definately does
2. 65nm has 2 major benefits, cost/chip and (in this case) power reduction. However, most people forget that in the case of cost, yields are an even bigger factor. AMD is currently enjoying over 85% yield on their 90nm process (an exceptionally high yield). By delaying their volume production of 65nm while simultaneously improving their yields on it, AMD is able to optimize their return for the process. In fact, they have a superb bit of software (called APM) that calculates the best time for the transition.
3. AMD did exactly the same thing with 90nm...they began volume production at close to mature yields. They have stated that 65nm volume production will START at already mature yields...

still they will not reach that point because I still feal that in many people eyes (though the number gets smaller everyday), they are a second rate product.

I agree that this is the biggest hurdle (perception). The only upside is that AMD is now being pushed by every OEM except Dell (and who knows when/if they will convert). The key indicator here is that AMD has gone from 6 Fortune 100 companies as clients to over 90 in the last 2 years. Consumers are a very small part of the equation...

1) May I have a source for these numbers? I'd also like to know how the 85% yield is measured (85% of chips will function at 100 MHz or 85% of chips will function at 2 GHz? BIG difference). I'm also quite sure that Intel runs very thorough analyses to determine the optimal transitions from one process to the other; it's not like they decide these things.

2) AMD has a very real need to release 65nm ASAP. I don't know why people refuse to believe that process is king in high volume ICs. Performance-wise they may not need it, but you can bet your ass that 65nm a die coming off a 300mm costs less for Intel to ship than a 90nm die costs off a 200mm wafer costs AMD. In the end it's all about money; and that means that AMD needs 65nm ASAP to remain competitive.

3) AMD are effectively a year behind in process technology, Intel was ramping 65nm production in 2H '05 and had volume shipments of 65nm in Q1 '06. AMD says they'll have production 65nm in 2H '06 and will ship in volume in Q1 '07. Keep in mind that Intel volume shipments are still significantly higher than AMD volumes.

AMD makes a great product at the moment but they need to seriously get their ass into gear as far as process technology goes or Intel will eat them alive sooner or later.

 

[Furen]

Originally posted by: openwheelformula1
"The key indicator here is that AMD has gone from 6 Fortune 100 companies as clients to over 90 in the last 2 years."

I was not aware of this great achievement. Wow.

That kind of statistic means nothing. If you could say that AMD has 20% of the business of the Fortune 100 then that's something but saying that 90 Fortune 100 companies could mean that they all bought one server and are using Xeons for the rest.

 

[Furen]

Originally posted by: RaynorWolfcastle

2) AMD has a very real need to release 65nm ASAP. I don't know why people refuse to believe that process is king in high volume ICs. Performance-wise they may not need it, but you can bet your ass that 65nm a die coming off a 300mm costs less for Intel to ship than a 90nm die costs off a 200mm wafer costs AMD. In the end it's all about money; and that means that AMD needs 65nm ASAP to remain competitive.

Going to a new process requires an upfront investment in order to save cash over the process' life. Intel has the cash to throw left and right while AMD has to manage itself a bit more carefully. AMD needs 65nm for higher clocks and lower power consumption, not for cost (though having lower costs is always nice). Capacity rules comoditized IC markets but CPUs are hardly comoditized, AMD can handle a time with an inferior process, just like it always has. Having fab36, in fact, will make this even easier as it will not be capacity constrained, remember those huge 200 sq. mm. hammers that AMD had to make on its 130nm SOI process with only one fab?

 

[RaynorWolfcastle]

Originally posted by: Furen

Originally posted by: RaynorWolfcastle

2) AMD has a very real need to release 65nm ASAP. I don't know why people refuse to believe that process is king in high volume ICs. Performance-wise they may not need it, but you can bet your ass that 65nm a die coming off a 300mm costs less for Intel to ship than a 90nm die costs off a 200mm wafer costs AMD. In the end it's all about money; and that means that AMD needs 65nm ASAP to remain competitive.

Going to a new process requires an upfront investment in order to save cash over the process' life. Intel has the cash to throw left and right while AMD has to manage itself a bit more carefully. AMD needs 65nm for higher clocks and lower power consumption, not for cost (though having lower costs is always nice). Capacity rules comoditized IC markets but CPUs are hardly comoditized, AMD can handle a time with an inferior process, just like it always has. Having fab36, in fact, will make this even easier as it will not be capacity constrained, remember those huge 200 sq. mm. hammers that AMD had to make on its 130nm SOI process with only one fab?

My point remains, it's not sustainable in the long run to fall behind on process technology to the competition. The negative impact of AMDs process lag in the past couple of years has been mitigated by Intel's inability to deliver on architecture. Assuming that Merom/Conroe is a solid architectural basis for a few years, AMD will have serious issues if it continues to lag on process.

 

[Viditor]

Originally posted by: RaynorWolfcastle

Originally posted by: Viditor

Intel had 65nm volume production in 05, so about a year behind (probably less)

About 6 months if you measure by release dates...but you have to keep in mind a few things:
1. AMD has no need to release 65nm, while Intel definately does
2. 65nm has 2 major benefits, cost/chip and (in this case) power reduction. However, most people forget that in the case of cost, yields are an even bigger factor. AMD is currently enjoying over 85% yield on their 90nm process (an exceptionally high yield). By delaying their volume production of 65nm while simultaneously improving their yields on it, AMD is able to optimize their return for the process. In fact, they have a superb bit of software (called APM) that calculates the best time for the transition.
3. AMD did exactly the same thing with 90nm...they began volume production at close to mature yields. They have stated that 65nm volume production will START at already mature yields...

still they will not reach that point because I still feal that in many people eyes (though the number gets smaller everyday), they are a second rate product.

I agree that this is the biggest hurdle (perception). The only upside is that AMD is now being pushed by every OEM except Dell (and who knows when/if they will convert). The key indicator here is that AMD has gone from 6 Fortune 100 companies as clients to over 90 in the last 2 years. Consumers are a very small part of the equation...

1) May I have a source for these numbers? I'd also like to know how the 85% yield is measured (85% of chips will function at 100 MHz or 85% of chips will function at 2 GHz? BIG difference). I'm also quite sure that Intel runs very thorough analyses to determine the optimal transitions from one process to the other; it's not like they decide these things.

Well, they come from a confidential analyst's report that I've seen (sorry...I wouldn't blame you for any skepticism). The number was calculated based on overall sales and on capacity for production. Keep in mind that AMD burned through all inventory, making this easier to calculate than usual...it represents an average yield for all processors (i.e. it's probable that some had a 75% yield and others a 95% yield).
I'm sure Intel also runs through optimization of it's transitions, but the 2 situations are not equivalent. The market share shift from Q4 is a perfect example of just how badly Intel needs to reduce power consumption and increase performance on it's chips. Even if their 65nm yields are low at the moment, the need to release 65nm ASAP is paramount to them.

2) AMD has a very real need to release 65nm ASAP. I don't know why people refuse to believe that process is king in high volume ICs. Performance-wise they may not need it, but you can bet your ass that 65nm a die coming off a 300mm costs less for Intel to ship than a 90nm die costs off a 200mm wafer costs AMD. In the end it's all about money; and that means that AMD needs 65nm ASAP to remain competitive.

I think if you restate that to be "process with good yields" is king, I'd agree. Keep in mind that you are talking about 2 different processes here as well...300mm and 65nm. AMD is already ramping volume production on 300mm at 2 Fabs (should be shipping for revenue by Q2). The shift to 65nm really isn't necessary to AMD from a functional standpoint, and obviously their APM software is telling them that it isn't yet a good idea from a financial one either.

3) AMD are effectively a year behind in process technology, Intel was ramping 65nm production in 2H '05 and had volume shipments of 65nm in Q1 '06. AMD says they'll have production 65nm in 2H '06 and will ship in volume in Q1 '07. Keep in mind that Intel volume shipments are still significantly higher than AMD volumes.

Actually, what they said is that they will have DESKTOP volume shipments in Q1 07...it is widely believed that these will be the last ones shipped (i.e. server and mobile 65nm will be the first chips delivered). The only other indication Hector made in the conference call was "we have already microprocessor products build on 65 nanometer that are really looking as planned"

AMD makes a great product at the moment but they need to seriously get their ass into gear as far as process technology goes or Intel will eat them alive sooner or later.

I really don't think (NGMA hype aside) Intel will be a major threat until at least 2008 (maybe late 2007). That's not to say that AMD will keep taking massive amounts of marketshare like they have been, but I don't think Intel will be able to take any marketshare back either...

 

[Cooler]

So soon AMD will reach intel highest clock speed in year with the move to 65nm
Seem the FX-64 (Dual Core) will be at 3Ghz.

If intels next gen is bust then Intel will be really hurting.