The AMD Execution Thread [2007 - 2017]
- Thread starter overclocked_enthusiasm
- Start date
-
- Tags
- amd
- Status
first of all, some of you seem to think i am making things up about SMIC. OK, it is my own analysis, but others think the same - it is a fringe idea and i will post some links now for you:
http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=205208561&pgno=2
http://www.fabtech.org/content/view/3193/120/
http://www.fabtech.org/content/view/5930/
http://www.eetasia.com/ART_8800495519_480200_NT_d86e6414.HTM
http://www.crm-daily.com/story.xhtml?story_id=100008D3CRMK
http://www.theinquirer.net/gb/inquirer/news/2008/05/07/smic-won-jump-450mm-bandwagon
Frankly, i think i make a good case an i could go on. But i promised i wouldn't. But i am not insane anymore, i think.
Alright apoppin, you wanted a response about AMD and SMIC. If SMIC can offer IBM's 45nm SOI process and if they won't be lagging behind AMD in terms of process transition, then I suppose AMD could outsource a part of their production to SMIC if the need arises. Then perhaps AMD will cancel their New York fab plans. But remeber this is all just a bunch of ifs and maybes.
Probability that SMIC doesn't substantially lag behind AMD for their 45nm transition: MINUS INFINITY. Regarding a strategic investor being interested in fabs from the IBM Alliance, I did hear some things on that matter recently, but honestly that doesn't create any direct link between AMD and SMIC.
IMO, appopin, your scenario isn't based on facts or likely events; you are merely misunderstanding fab and process technology dynamics. Now, since you seem to be in a pro-NVIDIA mood, I will assume that you couldn't possibly resist listening to a 90 minutes-long conversation between Jen-Hsun Huang, CEO/Founder of NVIDIA and Morris Chang, CEO/Founder of TSMC. It won't put to rest all of your ideas, but I suspect it might help a bit; plus, it's a nice conversation no matter what: http://youtube.com/watch?v=u-x7PdnvCyI
IMO, appopin, your scenario isn't based on facts or likely events; you are merely misunderstanding fab and process technology dynamics. Now, since you seem to be in a pro-NVIDIA mood, I will assume that you couldn't possibly resist listening to a 90 minutes-long conversation between Jen-Hsun Huang, CEO/Founder of NVIDIA and Morris Chang, CEO/Founder of TSMC. It won't put to rest all of your ideas, but I suspect it might help a bit; plus, it's a nice conversation no matter what: http://youtube.com/watch?v=u-x7PdnvCyI
Thank-you for the Youtube video link. i will d/l it today and listen to it tonight or in my car on the way to work tomorrow AM; too bad there isn't a transcript; i hate watching or listening when i can read. =P
.. Perhaps i am misunderstanding it just as are there are some other fringe analysts that believe what i do; it doesn't matter in my scenario that SMIC is behind as AMD as lags anyway in migrating to 45 nm and 65 will do for now. We'll see of course, what AMD actually does; it appears that they have more choice than the doom and gloom this thread allows them. Personally, although i now prefer nVidia, i believe AMD will survive .. and prosper.
No i can't back it up .. just a feeling that AMD has a clear direction and they have learned to be even more clever in hiding their plans from us. But then it is a "we'll see" and i love that this thread will continue to follow them. However, unless there is something really new, i will let it go and not keep on pushing for my idea as the only likely one. You guys are probably right .. probability is on your side. i am just looking at the unexpected; that is my own function at any forum.
But SMIC's 65nm process isn't from the IBM alliance so it'd be even more effort to port to it, and I haven't seen any indication that the process is actually any good. I'd much rather move to TSMC's 40nm process with more than 3x the perf/mm² rather to go to SMIC's 65nm process and have to migrate again 6-12 months later to not even catch up with TSMC's process density-wise...
Oh, on that front I certainly am not going to disagree with you - they're certainly doing a good job at hiding their manufacturing strategy. They're also being a bit slow to implement it apparently, but heh.
Well, it can be interesting to look at the unexpected; however then it might be a better idea to investigate multiple unlikely possibilities. Basically, if the most likely scenario has a ~60% probability, and there are 4 other scenarios with 10% probability... Ranting on and on about a single one of those scenarios is fairly ridiculous, but researching and explaining the pluses and minushes of all four unlikely scenarios can definitely be interesting and insightful.
But SMIC's 65nm process isn't from the IBM alliance so it'd be even more effort to port to it, and I haven't seen any indication that the process is actually any good. I'd much rather move to TSMC's 40nm process with more than 3x the perf/mm² rather to go to SMIC's 65nm process and have to migrate again 6-12 months later to not even catch up with TSMC's process density-wise...
Oh, on that front I certainly am not going to disagree with you - they're certainly doing a good job at hiding their manufacturing strategy. They're also being a bit slow to implement it apparently, but heh.
Well, it can be interesting to look at the unexpected; however then it might be a better idea to investigate multiple unlikely possibilities. Basically, if the most likely scenario has a ~60% probability, and there are 4 other scenarios with 10% probability... Ranting on and on about a single one of those scenarios is fairly ridiculous, but researching and explaining the pluses and minushes of all four unlikely scenarios can definitely be interesting and insightful.
You guys are the awesomest; i am really researching now and you have given me the unexpected!! Perhaps ammo to blow your own argument away, now i only listened to part of the youtube video and here is what i am able to transcribe [more-or-less]:
The very first question asked from the audience - check 1:08 in the video - got a very interesting response. The section was about innovation and fabs in the USA feeling they cannot keep up the pace any more (TI in particular), the founder of TSMC said in reply, that Intel was still around and keeping up well and also said,
--Here is the key imo:
The above is a very interesting point from someone who should know, being the founder of TSMC, the dominant semiconductor force in their region and now he is watching his fab's competition start to really emerge. I'm quite sure AMD is not blind to any of this either. They cannot afford a China fab of their own, but they can partner with someone already there, and AMD as a company prefers to do things via partnerships; we can see this with ATi and other partnerships they form.
Morris Chang (the founder of TSMC) is well placed to answer questions on Texas Instruments because he worked for them in quite an important capacity starting back in the 50's and it is basically how he developed his know-how to start TSMC.
Just because TSMC has always dominated Southern Asia, doesn't mean that they always will - especially now China is showing an interest. TSMC will probably become the Asian intel, but that doesn't mean there isn't plenty of room for others to compete, and thats what you guys apparently aren't getting.
Now, the argument here is saying that a lot of SMIC's fab processes are old, and using that as a reason trying to say SMIC makes no sense for AMD. While that old processes may be true, it totally ignores that SMIC has mostly manufactured memory products up until now which do not require advanced processes; so it really means nothing in regard to what AMD & SMIC might do together now.
Finally Arun, you commented about the SMIC 65nm process likely being different to AMD's. Clearly. imo, This is where ATI's engineers get to either prove their worth to AMD or resign themselves to stuff like fusion. ATi engineers are used to working with different FABS (TSMC, Chartered, UMC) and processes in the manufacturing of GPU's and they do so very quickly!! GPU architecture ages quickly, as we are well aware. These engineers should be able to help AMD port their CPU's between FABs and processes if the merger really is/was worth anything more to AMD than just a graphics sideline (i hope).
No, the opposition is just completely worthless.
Random permutations of story lines rarely exceed the level of garbage. This is no exception.
I could write a whole novel about how things really work, but allow me to an appeal to non-authority: over the years, I've worked on processes from 1um all the way down, produced in fabs of 7 different companies. By your standards, I'd be the perfect guy to consult AMD out of their misery. Yet I wouldn't know a TSMC process from a UMC one under an electron microscope.
The whole idea about standard cell design, as practiced by ATI and pretty much everybody else, is move up in abstraction to avoid the whole mess about process details: they simply don't need to know, it's handled by the library that's provided by the fab.
On the other hand, CPU designs with custom logic are intimately tied to a particular process. The last thing you want to do is abstract those details away. It's a completely different kind of knowledge. AMD has hundreds of engineers who know. ATI doesn't need any (or very few of them.)
If it's your ambition to delight us with your 'analysis', is it really too much to ask to limit yourself to topics about which at least know something?
You again dismiss my analysis without even understanding it; nor do you have any argument to counter mine, except to very rudely say i am stupid. Perhaps the opposition has much more worth than you realize. You are not always right.
You also completely ignore that if asset-lite is going to work - i.e. AMD gives up their fabs - then they are going to HAVE to go from the custom AMD design rules and process to whatever the destination fab uses. AMD engineers only have experience with Chartered there. That was the whole point of what I wrote before. Basically, AMD will have to compile their existing designs for a generic process, and then adapt that to the fab the chip will be fabbed in.
i am done here and you can believe what you guys want to. My minority report is right here for all to see - or ignore - and i will remind you 'what happened' in the future. Then we will see who was talking trash then, or not. Until then, you are clearly also guessing and have nothing concrete to reply to me - except ridicule. Carry on with out me.
Aloha
I really dislike the common misconception about AMD's supposed upcoming "asset lite" strategy, which many have taken to mean "fabless".
If you believe this to be the case, please read the following:
AMD's x86 co-licensing agreement with Intel precludes them from giving up the entirety of the manufacturing of x86 chips over to any other entity, including any other company. AMD is actually required by the details of this agreement to manufacture most of their own x86 microprocessors.
Any speculation which contradicts these fundamental principles which govern AMD's ability to manufacture and sell x86 microprocessors, cannot be true.
Geez apoppin,
Youve been doing this sort of posting and disagreeing all over the B3D boards in different threads and subjects. Ive noticed that you are not making (m)any friends around here with the people who have earned their respect as being the most knowledgeable.
Heres a hint, if the large majority of the most learned people say that what you are saying is rubbish, or at least has a VERY high chance of being rubbish....Chances are it is.
Just out of interest apoppin what is your main point of interest? With respect to technology/3D what do you most enjoy and follow?
lol, thats a bit harsh !
Youve been doing this sort of posting and disagreeing all over the B3D boards in different threads and subjects. Ive noticed that you are not making (m)any friends around here with the people who have earned their respect as being the most knowledgeable.
Heres a hint, if the large majority of the most learned people say that what you are saying is rubbish, or at least has a VERY high chance of being rubbish....Chances are it is.
Just out of interest apoppin what is your main point of interest? With respect to technology/3D what do you most enjoy and follow?
lol, thats a bit harsh !
You intend to drag me back into this thread, don't you? After i promised to leave! Look, i post in two threads in B3D ... 2!! That is two active threads!
- it would have been down to one if you did not also exaggerate and call me by name! Where is my "disagreeing all over the B3D boards in different threads and subjects" ?
As to your so called "experts", they would not last One Hour at ATF video or ARStechnica. Their arguments would be cut to shreds and they would be driven out of there. People are comfortable with their home forum and rarely leave like i did.
Look, i am not stupid, after moderating for them, Derek and Anand invited me to be an Editor for AT's main site. i felt i was not ready and i refused, OK?; plus i don't like their attitude very much so i also quit supporting ATF video; i just check my PMs there. i was the undisputed Alpha Male at ATF video for nearly the last 3 years and i also have important industry contacts far beyond simple focus groups. i also cannot reveal what i know and i cannot blab even though i do not sign any NDAs for any company. i guess they contact me because i never give up my contacts or embarrass them.
Now, this is a NO WIN situation for me here at B3D; if i don't say anything about myself, i am called stupid - if i tell the truth; that i am a writer and an have inside contacts, i am accused of bragging and having a huge ego. Now, what shall i do? You guys use my reputation against me; Either way. You can see the rudeness to me here and there is closed-mindedness everywhere. I've just reread silent guys rude reply to me.
He's ignoring that if asset-lite is going to work - i.e. AMD gives up their fabs - then they are going to HAVE to go from the custom AMD design rules and process to whatever the destination fab uses. AMD engineers only have experience with Chartered there. That was the whole point of what I wrote before. Basically, AMD will have to compile their existing designs for a generic process, and then adapt that to the fab the chip will be fabbed in.
Then i saw the his reply - again ignoring my argument completely - and stating that asset lite cannot mean AMD selling its own fabs.
Frankly, I disagree. It is my right to disagree with him, he is not on AMD's board! If the alternative to selling their fabs is going out of business altogether then I think AMD will be allowed to sell its fabs.
Intel won't really have much choice -contractual agreement or not but to allow this to happen - unless it wants the US Justice Department (and probably the EU also) to descend upon intel, with a nice juicy anti-monopoly investigation. I feel this will be especially true once Bush leaves the White House and the Democrats take over.
This is my take on how things would play out. My take on it. And if i am somehow right - as other HW analysts i linked to, now agree - then where are your experts and their majority opinion? it is opinion on both sides. Mine is the minority report.
You asked me about my interests; i am a writer, i am writing my own unrelated book currently; i am an investigative reporter, and my interests are the HW industry, specializing in discreet graphics. i AM very interested in Benchmarking - i found a way to save 90% of the traditional time involved in side-by-side IQ comparisons and was working on an article comparing r600 and g80's IQ over their evolving drivers over the past 12 months. Well, now i am not doing the article [mostly Crysis and BioShock] but DW of AT now knows basically how i do the time saving benchmarking; so watch for it there under another author; not mine. i am here at B3D simply because you guys have the best Benchmarking articles and good insights into the industry. And i simply want to learn; i am not ready to be a HW editor; i am pretty sure of that. But the active threads on benchmarking here at B3D forum do not currently interest me or apply to what i am doing, so i am hanging out in only two B3D "speculation" threads - formerly my own specialty at ATF video.
EDIT: i m equal opportunity; i search and i also link for everyone to see. Here is something to support the argument against me:
http://www.tgdaily.com/content/view/37423/118/
However, i am still confident in my PoV as i did expect that AMD would initially use TSMC as a foundry; the real question is though, will they continue to use TSMC for the long haul if SMIC can provide a cheaper alternative?
Last edited by a moderator:
What exactly is SMIC's future?
As you said, it's moving from memory fab processes, most likely because it is a bit player in a market that is suffering from a stupendous glut in manufacturing capacity brought about by a lot of very big players.
SMIC seems to be the first one to blink in an industry-wide game of chicken.
By licensing IBM's 45nm process, SMIC is moving into a field that is populated by a number large players, though the talk in the foundry business of issues with capacity driving down per-wafer revenue is nowhere near that of the memory sector.
SMIC is going to be late to 45nm, in relation to pretty much every other foundry and even AMD, it seems.
It doesn't seem to have the same clout, nor does it have the same long-standing relationships with the ATI design group when it comes to fabricating GPUs, the apparently dominant element to the first instantiation of Fusion.
Could SMIC undercut TSMC? I'm not familiar enough with the foundry dynamics to say, though I'd suspect it could not do so without seriously undercutting its R&D.
The process treadmill is turning into a cash-burning game, which is why TSMC is doing so well compared to IBM, AMD, and friends.
The IBM process alliance is heavily dependent on a number of unknowns, such as IBM's willingness to plow more resources into its microelectronics division in the long term, and AMD's long-term existence.
TSMC's cash flow is far more certain, and its volume is pretty much a given. Its transitions have proven thus far to be more timely as a result.
Can SMIC offer the same? Its starting point is a belated entry into a laggard process alliance.
Could it offer a cheap alternative at a competitive node in a timely fashion?
What data indicates it can be that cheap?
What kinds of cost can it cut?
Cheap labor isn't enough.
As you said, it's moving from memory fab processes, most likely because it is a bit player in a market that is suffering from a stupendous glut in manufacturing capacity brought about by a lot of very big players.
SMIC seems to be the first one to blink in an industry-wide game of chicken.
By licensing IBM's 45nm process, SMIC is moving into a field that is populated by a number large players, though the talk in the foundry business of issues with capacity driving down per-wafer revenue is nowhere near that of the memory sector.
SMIC is going to be late to 45nm, in relation to pretty much every other foundry and even AMD, it seems.
It doesn't seem to have the same clout, nor does it have the same long-standing relationships with the ATI design group when it comes to fabricating GPUs, the apparently dominant element to the first instantiation of Fusion.
Could SMIC undercut TSMC? I'm not familiar enough with the foundry dynamics to say, though I'd suspect it could not do so without seriously undercutting its R&D.
The process treadmill is turning into a cash-burning game, which is why TSMC is doing so well compared to IBM, AMD, and friends.
The IBM process alliance is heavily dependent on a number of unknowns, such as IBM's willingness to plow more resources into its microelectronics division in the long term, and AMD's long-term existence.
TSMC's cash flow is far more certain, and its volume is pretty much a given. Its transitions have proven thus far to be more timely as a result.
Can SMIC offer the same? Its starting point is a belated entry into a laggard process alliance.
Could it offer a cheap alternative at a competitive node in a timely fashion?
What data indicates it can be that cheap?
What kinds of cost can it cut?
Cheap labor isn't enough.
For the last time:
AMD HAS TO MANUFACTURE MOST OF THE X86 CHIPS THEY SELL OR THEY CAN'T SELL THEM AT ALL!
AMD HAS TO MANUFACTURE MOST OF THE X86 CHIPS THEY SELL OR THEY CAN'T SELL THEM AT ALL!
Last edited by a moderator:
Just as a general commentary, I don't think we need to call individual posters out on style or need to compare any given site to any other.
We can debate the message and the ideas.
I don't see a need to heat things up with debate of anything outside of the main topics.
We can debate the message and the ideas.
I don't see a need to heat things up with debate of anything outside of the main topics.
You're right, there isn't any real need for it. I'm sure someone will make a "noise" comment soon though
You're right, there isn't any real need for it. I'm sure someone will make a "noise" comment soon though
Yes, you just did, thanks. From my own link i posted, i find it interesting that TSMC has added SOI tech to their fabs. It certainly puts a new spin on things for me; however i still feel that AMD would be foolish to limit themselves to just one third part fab. i believe that China will start to grow strongly when it comes to semiconductor fabrication. TSMC has been lucky in the past that no other established or emerging fab has really been interested in their own business model - that SMIC appears to be copying right now!.
Never forget, the Chinese are masters of imitation and I expect they want to imitate the success TSMC has had in their own "rebel province" (as China thinks of Taiwan). It is what China does by undercutting their competition and they are doing whatever it takes. i am simply pointing that one thing out. Just one thing that is different from your traditional analysis. Of course i can be wrong; so can you, if you completely ignore SMIC, imo. i also agree with you AMD about using TSMC now - clearly i agree; i just have an added insight, i think. Is that so hard for you to take to take? .. a single prediction that other analysts i linked to also share?
AMD had to sell their fricking x86 chips before; no more! imo if they do not change to a "true asset-lite" they will simply die. You do not have to shout at me in all caps! And i am very sorry i confuse posters that attack me in the same manner, with each other. i will be more careful in future. Please, the message, not the poster!
Last edited by a moderator:
who is attacking you, apoppin? and even if someone is, so what? water off a duck's back, man.
how do you justify AMD's existence as a fabless semi house that can't sell x86 chips? they're back to being ATi then!
how do you justify AMD's existence as a fabless semi house that can't sell x86 chips? they're back to being ATi then!
Apoppin, it would seem to me that if SMIC is just starting out on a path right now that's conducive to AMD farming out manufacturing to them, then it'll be at least 2-3 years where they'll be at a point where it would be worthless to farm out production to them. I suppose it could still happen, but AMD really needs process level parity with Intel to be competitive (or a major architectural advantage like in the past), and I don't see SMIC achieving that.
Nobody else is either, but at least others are close and their roadmaps keep them close, as well as having the resources to transition away from current manufacturing techniques to new ones in about 5 years time. If I had to make a wild guess, I'd say AMD's future is more dependent on partnering with whoever can keep process shrinks going past the 22nm level at a ready pace, and their choices in the past have already decided the next couple years ahead. There's no magic fab savior for AMD right now, nothing that will give them the market leading product they need.
Nobody else is either, but at least others are close and their roadmaps keep them close, as well as having the resources to transition away from current manufacturing techniques to new ones in about 5 years time. If I had to make a wild guess, I'd say AMD's future is more dependent on partnering with whoever can keep process shrinks going past the 22nm level at a ready pace, and their choices in the past have already decided the next couple years ahead. There's no magic fab savior for AMD right now, nothing that will give them the market leading product they need.
My prediction of the magic solution for this problem:
the US government, in a bid to keep the flailing American chip company afloat, will split Intel's manufacturing and development houses into separate companies, making the manufacturing arm an open foundry which anyone can use.
This will allow AMD to finally flex some muscle and the x86 CPU market will see the greatest resurgence ever.
Timeframe: 3-5 years
the US government, in a bid to keep the flailing American chip company afloat, will split Intel's manufacturing and development houses into separate companies, making the manufacturing arm an open foundry which anyone can use.
This will allow AMD to finally flex some muscle and the x86 CPU market will see the greatest resurgence ever.
Timeframe: 3-5 years
- Status
