Sign in with
Sign up | Sign in
Your question

AMDs Fab 36 producing revenue

Tags:
Last response: in CPUs
Share
April 4, 2006 6:51:51 PM

I thought this was a good article. Nice to see AMD moving up and expanding. Competition is good and looks like AMD is on track for 65nm production later this year.


http://www.dailytech.com/article.aspx?newsid=1613
April 4, 2006 6:58:24 PM

Quote:
I thought this was a good article. Nice to see AMD moving up and expanding. Competition is good and looks like AMD is on track for 65nm production later this year.


http://www.dailytech.com/article.aspx?newsid=1613


Hmm....very interesting, especially the talk about "Good Competition"....

~~Mad Mod Mike, pimpin' the world 1 rig at a time
April 4, 2006 7:02:33 PM

AMD still has not shown working 65nm parts. No, wait, they have not shown 65nm non-working parts either. This looks very bad, Intel showed 45nm wafers a long time ago. When AMD puts out fluffy information-free calls without parts, you have to ask wonder how badly things are off in their processes and fabs.

It is gonna take some fancy tweaking if they want to break 3GHz by much on the current design.
Related resources
April 4, 2006 7:18:40 PM

Quote:
AMD still has not shown working 65nm parts. No, wait, they have not shown 65nm non-working parts either. This looks very bad, Intel showed 45nm wafers a long time ago. When AMD puts out fluffy information-free calls without parts, you have to ask wonder how badly things are off in their processes and fabs.

It is gonna take some fancy tweaking if they want to break 3GHz by much on the current design.


Hmm...I could care less if Intel showed off negative-nanometre chips, makes no difference, it only proves they NEED that to beat AMD, and it's the truth fanboy.

Do you want me to whip out the links? You're saying it's hard for them to break 3GHz by much? Look at Intel, they couldn't break 3.8GHz without melting the CPU, the only reason they can't break 3GHz by much, is because the ARCHITECTURE isn't built around GHz, the P4 is. If the Athlon 64 had 30 stages, it could EASILY reach 4GHz and beyond, but it's actually a GOOD architecture, that is more related to RISC than CISC and thus is why we have no 3GHz+ by much, but I'm sure your all-knowing self knew that already. :roll:

~~Mad Mod Mike, pimpin' the world 1 rig at a time
April 4, 2006 7:36:39 PM

Quote:
it only proves they NEED that to beat AMD, and it's the truth fanboy.


Quote:
but I'm sure your all-knowing self knew that already


Why does everything resort to name calling and insults with you?
April 4, 2006 7:37:02 PM

Quote:
Quote:
AMD still has not shown working 65nm parts. No, wait, they have not shown 65nm non-working parts either.


Maybe... but IBM & Chartered are already there.

This chart's from a year ago...

http://www.scfab.com/graphs/timeline_4.jpg

It's a pitty we don't get to see the rest of it...



Cheers!
April 4, 2006 7:38:23 PM

Quote:
AMD still has not shown working 65nm parts. No, wait, they have not shown 65nm non-working parts either. This looks very bad, Intel showed 45nm wafers a long time ago. When AMD puts out fluffy information-free calls without parts, you have to ask wonder how badly things are off in their processes and fabs.

It is gonna take some fancy tweaking if they want to break 3GHz by much on the current design.


Rich, you are a true troll and do not sound very intelligent. You sure are dissing AMD a lot which is weird considering they currently make the world's fastest microprocessor.

I'm sure AMD has a lot of things that Intel has, but one thing it does not have is a bloated advertising budget.

Yes, I believe Conroe will come close to living up to the hype. However, I also believe that AMD will not take it sitting down and will fire back with some hard punches.
April 4, 2006 8:00:16 PM

Well, I gotta be honest, I dissed AMD alot too. That was until I actually tried them and now, I love them. But one thing is, my thoughts have always been, unless you've actually tried them, don't diss them.

I dissed AMD until I built my first XP Athlon machine a few years ago. Now, I look back on it and realize how silly it was especially, now that i'm using them currently. Not to say I won't ever use Intel again because I still do.
April 4, 2006 8:04:41 PM

Quote:
AMD still has not shown working 65nm parts. No, wait, they have not shown 65nm non-working parts either. This looks very bad, Intel showed 45nm wafers a long time ago. When AMD puts out fluffy information-free calls without parts, you have to ask wonder how badly things are off in their processes and fabs.


It seems to me that AMD is waiting until they actually have a product ready to come to market to show it off. It doesn't do much good to show off the latest and greatest unless it's going to be available to buy.

Intel has made this mistake recently - they've shown off the Conroe, touting it's great benchmarks, etc - but can I go buy one off the shelf?? No.

I'd much rather wait to see a real product than to base hopes on one that only exists on paper and a hand-picked demo unit.

IDEV
April 4, 2006 8:09:09 PM

Quote:
You sure are dissing AMD a lot which is weird considering they currently make the world's fastest microprocessor.


That statement is completely relative.
April 4, 2006 8:34:54 PM

Quote:
it only proves they NEED that to beat AMD, and it's the truth fanboy.


Quote:
but I'm sure your all-knowing self knew that already


Why does everything resort to name calling and insults with you?

Because my mommy says I can.

Everybody is just believing this Intel Hype, and at the same time, saying "be neutral, competition is good, ya im cool". :roll:

~~Mad Mod Mike, pimpin' the world 1 rig at a time
April 4, 2006 8:53:06 PM

AMD has great production and I would like to find out about defects per 1 million. Does anyone have that info?
April 4, 2006 9:07:37 PM

I spent about 15 minutes googling... couldn't find anything new. Apparently the proper term is "defect density"
April 4, 2006 9:13:28 PM

Apparently defect density data is highly secretive. No one releases exact numbers on it...

Trade secret stuff I guess.
April 4, 2006 9:49:23 PM

Quote:
Apparently defect density data is highly secretive. No one releases exact numbers on it...

Trade secret stuff I guess.


They do share that data with suppliers - normally at the same time they are blaming that supplier for their problems! (Been there...)
April 4, 2006 9:59:24 PM

Quote:
AMD has great production and I would like to find out about defects per 1 million. Does anyone have that info?


Well, I don't have numbers but you can easily reach an average. AMD (and others) don't depend - exclusively - on their proprietary manufacturing assets: most lithographic processes' equipment is supplied by third parties (such as IBM, ASML NV, Nikon, Canon, ...). Although there are several very sophisticated techniques to reduce defect density, it depends on a variety of parametres and, to start with, with the quality of the wafers supplied.

You can find articles online on wafer manufacturing & lithographic processes but I guess all major chip manufacturers don't differ much, on output quality.
One of the biggest issues, is the node transition (in the current case, into the 65nm/300mm node): 65nm/300mm SoI wafers are much harder to procuce, in all aspects, than the previous 90nm/200mm (not to mention strained SoI wafers). Then again, the final quality is the same (take the AMD/Chartered/IBM example, for instance.). From the three big ones (IBM, AMD, Intel), only the later uses more "conventional" lithographic processes and relies, mostly, on "plain" strained Silicon.


Cheers
April 4, 2006 10:32:24 PM

TSMC claims to have a "near zero" defect density in its 300mm wafer facilities.

I wonder what they mean by "near zero"? LOL.
April 4, 2006 10:57:55 PM

AMD are progressing steadily here. It is inevitable that a company with significantly larger manufacturing caps are going to be able to whizz ahead to 45um, but we all know that qualifiying 'the best CPU' is complicated.

Does any one think that having the smaller process makes for a better CPU?

Current 90um AMDs are still performing very well against 65um intels.
I don't see that there is any need for AMD to be on the same node as intel.

If 65um AMD chips still perform well against intel chips @ 45um (including power consumption, raw speed etc) then what does it matter?
April 4, 2006 11:46:17 PM

Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.
April 4, 2006 11:58:13 PM

Quote:
Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.


You forget to mention that Intel also NEEDED 65nm for greater than 3.8GHz on the Netburst Single-Core and 3.2GHz Dual-Core. AMD isn't struggling with anything fanboy, Intel is the one gunning for 1st by going down nm after nm after nm just to keep up with a 3 year old architecture.

~~Mad Mod Mike, pimpin' the world 1 rig at a time
April 5, 2006 12:22:31 AM

Quote:
Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.


http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2734

Well, unless AMD has something 'big' up their sleeves (I doubt it), they'll have a hard time playing catch-up. BUT, performance-wise only, I don't think there'll be a huge difference. The problem remains, though: they're pushing a brilliant, outdated microarchitecture to its limits. That's the advantage of having a brand new one... working? :D 


Cheers!
April 5, 2006 12:41:39 AM

Quote:
Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.


You forget to mention that Intel also NEEDED 65nm for greater than 3.8GHz on the Netburst Single-Core and 3.2GHz Dual-Core. AMD isn't struggling with anything fanboy, Intel is the one gunning for 1st by going down nm after nm after nm just to keep up with a 3 year old architecture.

~~Mad Mod Mike, pimpin' the world 1 rig at a time

Well, your right only because Intels current architecture is an energy waste and moving to 65nm helped them make it more efficient. Now Intel can if they want, clock their chips higher but, we all know and they know netburst isn't working anymore. The biggest reason Intel moved to 65nm is to save production costs. It's the next logical step.

But you knew that already so, nuff said ..
April 5, 2006 12:52:45 AM

Quote:
TSMC claims to have a "near zero" defect density in its 300mm wafer facilities.

I wonder what they mean by "near zero"? LOL.


...'bout 3-per-wafer!
I guess you wouldn't be [that] surprised with what technology can achieve, nowadays...

http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do;jsessionid=EzPFEIw876083Gcg300tbWpvMqEWgnnvut544O24sDjSXGFNE90U!343224493!170403259!7101!7002#

...and, for a glimpse on how it works:

http://www.nikon.co.jp/main/eng/portfolio/about/technology/nikon_technology/immersion_e/index.htm


Cheers!
April 5, 2006 12:56:39 AM

I built away from the prosscer wars, away from knolege of a manufature, I just bought what was right, but then I left Canada and only buy AMD (still curently the best).
April 5, 2006 8:42:22 AM

Quote:
Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.


And??? AMD have already proven to us that high clock speed does not a fast processor make.

You have a point if AMD make no major revisions to their design, but we already know that K8L is around the corner.

AMD have increased performance per clock before, when I moved from a 1.73ghz Athlon to my current athlon 64 @ 1.8ghz, which is a small difference in clock speed, i noticed a huge jump in performance.
I think that there is no reason why they can't improve again, and not have to push higher clocks, like intel have been doing with netburst for the past few years.

In fact look at the Conroe core, that uses a much lower clock than AMD and gave higher benchmarks - What does that say about the importance of clock speed?
April 5, 2006 9:08:11 AM

Yes!

and Intel should thank AMD for taking the time and effort to prove to consumers that higher clock doesn't mean better CPU.

Now Intel is using the same weapon to arm it with Conroe.

Athlon64 with higher clock is not necessary a better CPU than lower clock Conroe!

Only future can prove it but Intel sure won't have to go on the same road as AMD did.
April 5, 2006 9:36:35 AM

That is because AMD already paved the road for Intel to ride on. I used to be a system builder/seller. And when AMD made the transition to the XP line it took a while to convince people that they were the better proc for the money. While Intel simply has to say "hey, you already know its possible because of AMD, so come buy our products".

One thing that you are missing here is that AMD is a smaller company then Intel. They simply do not have the resources like Intel does so yes it will take more time for them to go down on the nanometer scale. But, they are larger enough that it will be done. Also as many have said here before, Intel showed you the Conroe chip months before you can buy it. How big of a issue is that causing in the forums with bickering about how great it's gonna be? It has been my experience with AMD's products that with the exception of the A64, AMD doesn't tout around stuff month's and month's ahead of time. They do give some notice about certain things like the AM2 upgrade, but look at X2, not long after you knew about it, they were available. I am willing to believe that they are doing the same thing with the next gen chip which will most likely be on a 65nm process. The question simply is how long before we get it? As of right now, there is NO reason for AMD to say, "Hey, you might not wanna buy that new X2 chip cause the next new chip is just around the corner". They are the current leader in the CPU wars, so keep a couple cards under your sleeves. I am half expecting them to release a new chip architecture around the same time Conroe is expected to be released, just to stab at Intel. But that is simply my own personal theory, no info to base it on.
April 5, 2006 10:21:57 AM

According to road maps on HKEPC (not sure how-trustworthy)
AMDs next major architecture is the 65um Brisbane (dual) and sparta (quad) K8L cores.

http://www.hkepc.com/hwdb/am2-4800-1.htm

I haven't seen any hard and fast facts, but I'm sure we'll find out in time.
April 5, 2006 10:41:28 AM

Why did Intel choose to release Conroe sometimes around July or later?

- Enough time to produce enough chips to meet expect demand

- Kids are getting ready to go back to school or going to collgeges. Think about the number of average parents will be out looking to buy new pc systems and also learn about what's new with PC technology.

- College students are on summer break and it is a good time to make some quick cash for a new PC. Conroe is there as a good choice.

When else will it be a better time to try to sell some PCs? Not like everyone can afford a new PC as a Xmas gift.
April 5, 2006 11:06:44 AM

i was doing some searching and while it does not give an exact time, here is a quote from AMD's website concerning 65nm production:

Quote:
65nm
AMD is meeting its 65nm development goals. We have achieved our 65nm SRAM milestones, and our transistor and interconnect development is on schedule. We are on track to begin volume production on 65nm in 2006.

AMD's first 65nm transistor generation will take full advantage of our very latest 90nm high-performance SOI transistor technology, and will optimize this technology for our first 65nm products. This will enable AMD to achieve its aggressive 65nm product performance/power objectives, while helping ensure a fast and efficient manufacturing ramp.


Also in their 3 year roadmap it show 65nm production to begin in 2006. So AMD is not far behind Intel, it is just not tooting its horn about it.
April 5, 2006 11:09:03 AM

also in response to noblekitty, we are talking about processors here not game consoles. Processor manufactures are more likely to just release their products when they are ready as the bulk of their business comes from business customers who could care less about the time of the year it is. While you do present somewhat of a good point about the college thing, i highly doubt that has anything to do with the timing of release for next gen procs.
April 5, 2006 4:53:38 PM

Quote:
TSMC claims to have a "near zero" defect density in its 300mm wafer facilities.

I wonder what they mean by "near zero"? LOL.


This is a very funny statement, let's see what near zero really is.... the key word, if you quoted it correctly, is DENSITY. A defect density of 0.1 is near zero (0.1 defects/cm2) that equates to a total of 76 defects per wafer on 300 mm wafers :)  .... not quite to zero would you say.... now, at a die size of 100 mm2, this would average out to ~ 706 die at 76 defects per wafer, assuming random distribution, would come to around 10% of the die defected, this sound more like typical yields for a healthy fab. Hmmmmmmm..... sounds like a boast of mediocrity to me :) 

Quote:
But liquid mediums present their own challenges, including defects such as bubbles, watermarks, particles, particle-induced printing defects, and resist residue. TSMC’s R&D researchers resolved these issues by developing a proprietary defect-reduction technique that, on initial tests, produced less than seven immersion-induced defects on many 12-inch wafers, a defect density of 0.014/cm2. Some wafers have yielded defects as low as three per wafer, or 0.006/cm2. This compares to several hundred thousand defects produced by a prototype immersion scanner without these proprietary techniques and significantly better than published champion data in double digits.

TSMC’s immersion lithography technology is targeted at TSMC’s 45nm manufacturing process.

http://www.us.design-reuse.com/news/news12662.html

Well, these are TSMC's claims but, all things being equal and assuming a random distribution, wouldn't this give a bit over 1% defect density per die, instead of 10%? I'm not sure if they're referring to immersion-induced-only defects: if so, that would induce [us] in a major overall defect density error! :D 

(Of course, they do not disclose their «proprietary defect-reduction technique»; one can only infer an average according to current immersion lithographic techniques' yields...)


Cheers!
April 5, 2006 5:35:51 PM

Quote:


That said, I found Anand's article more enlightening:
AMD Fab 36 Press Release Summary

What mulls around in my head is one very nagging question that I cannot really make out what to think.....

Why on earth do you put your smallest die and lowest margin products (Athlon 64 and Sempron) on your largest, most expensive wafers from your brand spanking new flagship 300 mm fab?

The first die they sell starts the depreciation clock ticking, and each tick eats into their bottom line....it just does not make any sense.


Yes, that puzzled me too but I assumed that, since Fab30 is already yielding 30K (200mm) wafers-per-month and Fab36 is only expected to ramp up to the 20K (300mm) wafers-per-month mark in 2008, it made sense to me (and, not because «AMD cites customer demand as the reason...», nor «...there's always need for more capacity at the lower end,»). Another possibility: I might be totally wrong. :D 

Quote:
The parts that it is shipping are 90nm Athlon 64 and Sempron CPUs, so Opterons and Athlon 64 X2s will still come out of Fab 30 next door. AMD cites customer demand as the reason that Athlon 64 and Sempron are first out of the new fab, which honestly makes sense; there's always need for more capacity at the lower end.



Quote:
Although both Fab 30 and Fab 36 produce 90nm processors, Fab 30 uses smaller 200mm wafers while Fab 36 features an upgrade to 300mm wafers. AMD says that yields on 300mm wafers coming out of Fab 36 are comparable to 200mm yields being produced at Fab 30.



Quote:
In terms of capacity, by 2008 AMD plans to be able to crank out 20,000 (300mm) wafers per month out of Fab 36 which would double its microprocessor production capacity to approximately 100M CPUs per year (thanks to larger wafers and smaller transistors). For comparison, Fab 30 currently produces 30,000 (200mm) wafers per month. Unfortunately AMD was unable to provide us with any sort of guidance as to how quickly Fab 36 will ramp up to its 20K per month target.


Edit
: Moreover, prepping a node transition at the same time...
April 5, 2006 6:09:58 PM

You seem to know a lot about manufacturing technologies. Nice to have someone in here who does...

Anyways- I wanted to point something out to those who are arguing the technological advantages of die shrinks (specifically MadModMike, RichPLS, and others). I've said this before so I'll be brief.

One the most important things to a microprocessor company are yields. While shrinking dies certainly has both disadvantages and advantages, it's important to remember that there are NO disadvantages to die shrinks in terms of yields. AMD might be able to compete technology wise with 90nm on 300mm wafers against Intel's 65nm, but they CANNOT compete in terms of cost. This is where intel has a huge advantage and this is also why Intel continues to push for die shrinkage. In a climate of booming dual core processors, yields are more important that ever. Doing business on 90nm hurts AMD cost wise. They have to buy more 300mm wafers more than ever to keep up while Intel can probably keep buying the same number of 300mm wafers because they have higher yields w/ 65nm.

As I've said in the past, silicon wafers are very expensive. As far as actual costs of manufacturing a processor, wafer cost is the biggest. So the more dies you can fit on a wafer the less it's going to cost to make (per package).

So just keep that in mind guys. Die shrinking is often more about increasing yields than it is about increasing clock speeds or technological advances.

-mpjesse
April 5, 2006 7:00:35 PM

Quote:
You seem to know a lot about manufacturing technologies. Nice to have someone in here who does...

Anyways- I wanted to point something out to those who are arguing the technological advantages of die shrinks (specifically MadModMike, RichPLS, and others). I've said this before so I'll be brief.

One the most important things to a microprocessor company are yields. While shrinking dies certainly has both disadvantages and advantages, it's important to remember that there are NO disadvantages to die shrinks in terms of yields. AMD might be able to compete technology wise with 90nm on 300mm wafers against Intel's 65nm, but they CANNOT compete in terms of cost. This is where intel has a huge advantage and this is also why Intel continues to push for die shrinkage. In a climate of booming dual core processors, yields are more important that ever. Doing business on 90nm hurts AMD cost wise. They have to buy more 300mm wafers more than ever to keep up while Intel can probably keep buying the same number of 300mm wafers because they have higher yields w/ 65nm.

As I've said in the past, silicon wafers are very expensive. As far as actual costs of manufacturing a processor, wafer cost is the biggest. So the more dies you can fit on a wafer the less it's going to cost to make (per package).

So just keep that in mind guys. Die shrinking is often more about increasing yields than it is about increasing clock speeds or technological advances.

-mpjesse


Yeah, that's a good thought out post. That also explains why Intel (as far as we can see now) has cheaper prices on their Conroe chips coming out. Intel moved to 65nm for a reason and that's probably the biggest. They're pushing towards 45 and 32 and so on.

It is good that AMD is pushing toward 65 as well hopefully we'll see a better decrease in prices when that happens.

Good post mpjesse :wink:
April 5, 2006 11:59:44 PM

Quote:
You seem to know a lot about manufacturing technologies. Nice to have someone in here who does...

Anyways- I wanted to point something out to those who are arguing the technological advantages of die shrinks (specifically MadModMike, RichPLS, and others). I've said this before so I'll be brief.

One the most important things to a microprocessor company are yields. While shrinking dies certainly has both disadvantages and advantages, it's important to remember that there are NO disadvantages to die shrinks in terms of yields. AMD might be able to compete technology wise with 90nm on 300mm wafers against Intel's 65nm, but they CANNOT compete in terms of cost. This is where intel has a huge advantage and this is also why Intel continues to push for die shrinkage. In a climate of booming dual core processors, yields are more important that ever. Doing business on 90nm hurts AMD cost wise. They have to buy more 300mm wafers more than ever to keep up while Intel can probably keep buying the same number of 300mm wafers because they have higher yields w/ 65nm.

As I've said in the past, silicon wafers are very expensive. As far as actual costs of manufacturing a processor, wafer cost is the biggest. So the more dies you can fit on a wafer the less it's going to cost to make (per package).

So just keep that in mind guys. Die shrinking is often more about increasing yields than it is about increasing clock speeds or technological advances.

-mpjesse


:lol:  Sorry for laughing, but I'm no expert and I do not know a lot about manufacturing techs! I do make an effort to learn, that's all!

That said, I totally agree with you. And I think you've touched a very sensible point which summarizes most of what's been said on the subject.
Very well, indeed. :wink:


Cheers!
April 6, 2006 12:38:01 AM

something i read in the immersion lithography webpage said that they can go down to 45nm without much change to the production, one would assume that this statement means that Intel would have to do a complete revamp of its facilities to go down lower. I would think that may lock in the 45nm process for quite some time when they actually get to it.
April 6, 2006 12:54:12 PM

Quote:
Why on earth do you put your smallest die and lowest margin products (Athlon 64 and Sempron) on your largest, most expensive wafers from your brand spanking new flagship 300 mm fab?


300mm wafers have 225% more area than 200mm wafers. The costs for the wafers are not that much higher and the processing costs are not much higher either (they both go through same steps) but with a small die you can get a lot more full die around the edge of the wafer, thus increasing your Die Per Wafer as opposed to what you would get with larger die. Keep the larger die on the smaller wafers since your % increase in DPW is not as high for larger die.

In other words you could replace 20000 wafers starts of smaller die with ,say, 8000 wafers starts in your new fab and keep the 20000 wafers from other factories running bigger die. You end up making more chip that way.
April 6, 2006 3:28:35 PM

Quote:
Well they don't, you see. If AMD does not go to 65nm, then they have climaxed at 2.8GHz dual core/3GHz single core...
So 65nm is needed badly for AMD to shoot for the performance king again, and it looks like it will be a struggle considering the time lag getting there.


And??? AMD have already proven to us that high clock speed does not a fast processor make.

You have a point if AMD make no major revisions to their design, but we already know that K8L is around the corner.

AMD have increased performance per clock before, when I moved from a 1.73ghz Athlon to my current athlon 64 @ 1.8ghz, which is a small difference in clock speed, i noticed a huge jump in performance.
I think that there is no reason why they can't improve again, and not have to push higher clocks, like intel have been doing with netburst for the past few years.

In fact look at the Conroe core, that uses a much lower clock than AMD and gave higher benchmarks - What does that say about the importance of clock speed?

Intel's NetBurst's dead. But not [high] clock speeds.

In my opinion, we're nearing a confluence point, where improved & added features will meet high clock speeds.

Charles Severance (http://www.crew.umich.edu/people/cseverance.html) is supposed to have coined the term "post-RISC" architectures, when Both CISC & RISC ISAs began to take advantage of each other, and that made sense. But, it left a sort of void, in terms of a functional microarchitecture designation. Intel chose the "P6", the "NetBurst" & the "Core" designations; AMD chose the "Hammer" & the "K8x" (?). None translates into what CISC & RISC did, when things were that black & white.
Actually, more than just an interim feature-rich wise trend, microarchitectures seem to be following a new approach, a restart to higher frequencies through a different collection of technologies & techniques. Some keywords are 'new materials', 'top-to-bottom design integration' & 'new manufacturing processes', among many others.

Hence, instead of a single, paradigmatic "Moore's law", we're left with two more premises to be taken into account, the "Feature-rich" one and the pseudo-NetBurst "GHz ramp-up" one. The first deals with the increase in the transistor number per unit of time, the second with what can be done with that number per unit of time and the last, with how fast the first two can perform. Probably, all three will die at the same time. :D 

That's why Conroe will not start at 700 MHz...

Edit: This is, merely, my opinion. And, I'm not addressing Quantum Computing & the like! :wink:


Cheers!
April 6, 2006 4:37:53 PM

I've read/heard the same thing. Going below 45nm is supposedly going to take a major undertaking in both manufacturing technologies and cost. I've read that IBM has managed to figure out a way of getting below 45nm using current methods, but they've not yet demostrated it and it's not clear if it'll be feasible for full blown microprocessors.

I think from a technological standpoint die shrinks are going to become less important. The latest craze is multi die CPU's. From a business standpoint die shrinks will still be important 'cause we'll eventually be on 8 and 16 core CPU's. That eats a lot of silicon per package. ;-)

Who the hell knows? lol.
April 6, 2006 4:47:53 PM

Quote:
Anyways- I wanted to point something out to those who are arguing the technological advantages of die shrinks (specifically MadModMike, RichPLS, and others). I've said this before so I'll be brief.

So just keep that in mind guys. Die shrinking is often more about increasing yields than it is about increasing clock speeds or technological advances.

-mpjesse

It is about money, the die shrinking. Not about the yields, technology or good manners and human consciousness.
Before making an investment, an economical analysis is made. If there is logic and the income over certain period is bigger than the interest of bank-saving the money, than the investment is justifiable.

Chip production costs are much more depended of other determinants than the cost of the used wafers,like employees, equipment, marketing and other. Scientists, engeneers and technologists are needed for developing and inventing materials, technology, and architecture of the chips. Yes, this category costs a lot, same as the time and the experiments. During the production process workers, enginers and other employees are needed also. The used equipment(usefulness limited) - and buildings are taking a good part in the total cost also. Marketing is another important rate that goes with a part in the calculation.

So, the transition from 65nm to 90n costs money and needs time to be accomplished. AMDs math was wrong when they decided not to hurry with 65nm, like Intel did. They thought that is too early to invest and that their 90nm SOI K8s are good enough to compete with the new Intel Core chips. Now they know they are late and are rushing, not becouse of the yields, but becouse of the technological adventages that will let AMD to produce more competetive chips for the allready lost performance&price battle against Intel(Core architecture chips).
April 6, 2006 5:49:48 PM

i agree , AMD shouldnt make the same mistake as Intel a few years back when they took the " what the hell, our stuff is good nuff " and try sitting back and getting fat for a bit. didnt that cost intel a large share of the desktop market?? all in all , alot of people buy from advice handed down
from people like you and i (some knowledge of PC hardware).
I for one own 4 AMD systems in my home, bet your walnuts that will
change if Conroe isnt just Intel pumping smoke up our tails , AGAIN.
when it boils down to it , most people are gonna by the better CPU
( me , for one).
i hate the monster called Intel (but alot of rebels cheer for the underdog)
however, i'm gonna buy the FASTEST performance per dollar CPU out there. although these companies should beware of outsourcing to foreign
countries, companies like Dell and GM etc... are feeling the sting. people put up with slightly inferior products to support their friends and families.
when the company puts whole communities in the streets overnight to save some cash in mexico. how much do they profit when noone is left with enough cash to buy their product or willing to buy a slightly inferior product to support a country they hate? okay i strayed , SORRY =)
April 6, 2006 5:51:58 PM

Thanks for both reiterating what I already said and telling me something I already know.

Of course it's about money! I've said that 100 times in the past. That fact is implied in my post and it's why i said this at the beginning of my post:

Quote:
I've said this before so I'll be brief.


Higher yields per wafer = more money. Period. Making more money is the ultimate goal of any business.

And you're right, things like facilities, equipment, people, etc are part of the larger equation. However, things like equipment (probably the biggest expense) are a capital expeditures. Meaning that cost is not factored into the actual cost of the processor or product.
April 7, 2006 2:27:31 AM

Something also to highly consider is chip real estate. While multi core chips in theory sound cool. From the pics i have seen of a dual core AMD chip you could most likely only fit four cores on that thing without a major change in overall size to the chips meaning either one of two things. Die shrinkage to allow more chips in the same amount of space, OR the come back of slots, such as the P2 chip style connections to allow for the higher space. Now I am guessing that since Intel obviously decided a long time ago that the Slot architecture was not effective enough, that we are not gonna go back to that. Also i highly doubt that mobo manufacturers are gonna support highly compacting there stuff for oversized cpus. I mean seriously have you looked at some of the newer board, especially when you start looking at quad sli's and things, those things are already compacted enough. imagine having to make space for a proc 2-3-4 times as large as current day procs. as well as the amount of heat a proc of that size would produce. It would take gigantic heatsinks which also means a huge increase in weight against the board unless they are directly attached to the case underneath. Just my thoughts...
April 7, 2006 2:33:19 AM

bring on the debates




8O
April 7, 2006 4:13:49 AM

What comes after nanometer?
April 7, 2006 11:01:18 AM

A thought about that could be, that they are running the lower margin stuff on fab 36 Because its new. Think about it, why run your highest profit stuff on equipment that does not have years of proven experience when you have another fab that does have experience AND can produce what you need for that chips capabilities. Like it or not I am guessing that AMD sells lower end chips 2 to 1 over really high end chips. So while your margin maybe higher on the high end chips, if you don't have the low end chips to satisfy your budget minded consumers, as well as businesses trying to get by on the cheap, then there is a good chance they take their business to Intel.

Does anyone have any knowledge of AMD's sales to either back me up on this or prove me wrong. I just woke up and don't feel like googling it right now.
April 7, 2006 2:05:08 PM

Quote:
Say you get 100 CPUs from a 200 mm wafer, consequently you would get 225 CPUs from a 300 mm wafer. Now let's do the math

Semprons range in price: 60 bucks to 125 bucks (source: pricewatch.com)
Athlon 64's range in price: 134 to 804 (if we include the FX series) bucks.

So let's take the two highest prices units and get an ASP of 469 bucks per processor (this is way too high, but for the math, let's make it easy).

On 200 mm you will get: 46,900 bucks for 1 wafer
On 300 mm you will get: 105,525

For the large chips:

Opterons range in price of 128 to 1799 and X2s 295 to 627. Lets treat it the same way, average the two highest and we get an ASP of 1213. But because they are large, you can only fit 50 on a 200 mm wafer and 112 on a 300 mm wafer, therefore you get:
On 200 mm you will get: 60650
On 300 mm you will get: 135,856

It is simple, you will make 2.25 x more moving your larger die and higher margin material to 300mm.

I.e. the math is simple, you will get more money out of your 300 mm fab by working the highest margin products, plain and simple, you maximize your ASP by generating the parts that get the highest price. Now in this analysis, these are gross numbers and the ASP's are completely dumb, but the logistics are exactly the same. Why would you make your cheapest products, products that your competitor is driving to extinction such as single vs dual core, by the end of the year? It does not make sense.

Or does make sense, if you are not really yielding on 300 mm to make that 2.25 factor mean something, then I can understand why you would do this. Or, if you are yielding but the bin splits are not giving you the highest margin parts, then I can explain this behavior. See where this can go?

It is great that AMD is shipping product from Fab 36, I just question why they are shipping the lower grade products, and it cannot be because demand is sooooo high, even if it were, satisfying it is not really in their best interest. If raw volume is all they need, it makes more sense to try to satisfy that demand by freeing up capacity on your cheapest processing line. Why flood the market with oooodles of low price chips, you are cutting your own throat.



225% is the maximum theoretical usable area. you have to subtract the incomplete die around the edge. Larger die subtract A LOT from the 225% vs. the smaller die, and I am just guessing, but i am pretty sure that they only get a few high bin splits per wafer, so your averages are well off. The previous line you mention is likely to have the higher bin splits/wafer than the x2/opty's
!