Tom's Hardware > Forum > CPU & Components > CPUs > Is AMD going to split?

Is AMD going to split?

Forum CPU & Components : CPUs - Is AMD going to split?

Tom's Hardware: Over 1.4 million members in 6 different countries available to answer all your high-tech questions. Sign up now! Its free!
Word :    Username :           
 

Back a few months ago there was a rumor swirling around about AMD going to a FAB Lite strategy. Meaning they would offload their FABs and outsoruce their chip fabrication. Now we never had a 100% to this rumor but it is still around and has some interesting questions arising.

My questions are:

Is AMD planning a Asset Light and Asset Smart strategy?

How will this affect the CPU market?

How will this affect the GPU market?

Where does their ATI division stand with this decision if it does come true?

Will this benefit AMD or hurt them in the long run?

I am sure there are even more questions but these are some basic ones that come to my mind. I think the most important one is the last one. Being weather or not this is a good idea in the long run. ATI had and still uses a third praty FAB, TSMC, which worked out great for them. But will it work the same for AMD? And will it benefit AMDs margins to allow them to regain profitability?

Now we still do not know 100% if the rumor is true but one thing is for sure: Hector has not left. A lot of people believe that Hector ran AMD horribly. I myself think under his management AMD has suffered more than they have benefited. Back a few months ago as well Hector left the CEO position but did not leave his position as Chairman of the Board. Now this may not seem like anything big but as Chairman of the Board he has a lot of power and according to a post on THG he is going to stay there and try to see the Asset Light and Asset Smart strategy through:

Quote :

Hector Ruiz ’ time at the AMD as we know it today is coming to an end. Ruiz recently stepped down from his role as CEO, but kept his position as chairman of the board with the purpose of launching AMD’s Asset Light and Asset Smart strategies, which are expected to split AMD into two entities.


http://www.tomshardware.com/news/A [...] ,6109.html

Feel free to comment but please no flaming, stupidity or fanboyism. I am sure other people have questions and hopefully they will be answered in time. And hopefully this is a move for good not bad.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Sponsored Links
Register or log in to remove.
- 0 +

jimmysmitty wrote :

Back a few months ago there was a rumor swirling around about AMD going to a FAB Lite strategy. Meaning they would offload their FABs and outsoruce their chip fabrication. Now we never had a 100% to this rumor but it is still around and has some interesting questions arising.

My questions are:

Is AMD planning a Asset Light and Asset Smart strategy?

How will this affect the CPU market?

How will this affect the GPU market?

Where does their ATI division stand with this decision if it does come true?

Will this benefit AMD or hurt them in the long run?

I am sure there are even more questions but these are some basic ones that come to my mind. I think the most important one is the last one. Being weather or not this is a good idea in the long run. ATI had and still uses a third praty FAB, TSMC, which worked out great for them. But will it work the same for AMD? And will it benefit AMDs margins to allow them to regain profitability?

Now we still do not know 100% if the rumor is true but one thing is for sure: Hector has not left. A lot of people believe that Hector ran AMD horribly. I myself think under his management AMD has suffered more than they have benefited. Back a few months ago as well Hector left the CEO position but did not leave his position as Chairman of the Board. Now this may not seem like anything big but as Chairman of the Board he has a lot of power and according to a post on THG he is going to stay there and try to see the Asset Light and Asset Smart strategy through:

Quote :

Hector Ruiz ’ time at the AMD as we know it today is coming to an end. Ruiz recently stepped down from his role as CEO, but kept his position as chairman of the board with the purpose of launching AMD’s Asset Light and Asset Smart strategies, which are expected to split AMD into two entities.


http://www.tomshardware.com/news/A [...] ,6109.html

Feel free to comment but please no flaming, stupidity or fanboyism. I am sure other people have questions and hopefully they will be answered in time. And hopefully this is a move for good not bad.





Im not being fanboyish but I do find Intel is adding the final insult by announcing some new chips as news above the AMD Split, the timing was right at AMD's lowest ebb....... Highly amusing... I am really starting to feel for AMD now as I am installing proberly one of my last AMD chips a x2 5000 bought for £34........


It reminds me of a sketch where a woman is tied to a tree in a state distress and has been "attacked" by a group of men and found by a single male..

She is greatful to be "rescued", but is taunted with - I guess this aint your lucky day....

AMD is that lady tied to the tree


Reply to Hellboy
- 0 +

AMD will not "split into two". I grow weary of seeing all the dooms day prophocies for AMD. It is just going to be outsourcing some of its resources else where. That's it. Hopefully it will mean cheaper processor in AMD's new line come Q1 of 09.

As for the GPU market... I'm pretty sure it will have no effect at all. AMD "owns" ATI, the same way that the united states is a protectorit of puerto rico. Or something like that anyway.

Reply to E3210
- 0 +

Real men have fabs.

Reply to jeb1517

I really don't see how they can make that work as the Fabs would need more than just AMDs business considering that contract foundries are charged less. If you look at Chartered, AMD could get 100K chips from them for (TC) less than making them themselves.

How could the fab survive on that? And then you have to assume that they would just be trading Intel for TSMC as a competitor if they do try to make chips for others.

Moving fab of Radeon or chipsets would also be difficult as GPUs are larger and have less margin while chipsets are even worse (in terms of margins).

In other words, they would have to be trying to go out of business. But then, they did say that Dirk was misquoted and that he didn't say they were selling the fabs but were defining a revolutionary way of fabbing chips.

But then with the suicidal price war they got into who knows. The smartest thing in my mind is to push Stars cores to Griffin and pump the MHz. That would give them a 2.7GHz or so mobile chip (based on improvements noted from K10 - K10.5).

And of course, to get Opteron up to 3GHz which will pretty much kill Intel on the 4P side and perhaps even 2P considering the records they're setting with K10 only at 2.5GHz. Please no Nehalem comments. MP won't show up until late next year last I heard. By then AMD should have at least HighK if not preparing for metal gates. It was said earlier this year that they would implement HighK first (D0) and then metal gates (E0). Shanghai is C0.

So basically, it's highly doubtful that a spin off would help or is even planned.

Reply to BaronMatrix

I think the more interesting thing here that no one has commented on is the fact that Hector is still with AMD and in a very powerful position, and in fact a more powerful position than Dirk.

If Hector wishes to try this strategy he has a lot of power to push it through.

Now I think it has its up sides and down sides. The up side is of course lower cost for AMD thus resulting in better profits per chip and possibly lower cost for the end user (us). The downside is quality control will not be in AMDs hands and could cause potential backlash for the end users (us again).

As I said its still just a rumor but its been one of the longest running and I am interested to see if it does happen and if it truly does go through what the side affects will be.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty

You are more than likely turpit.

My only real question is how will this affect AMDs CPUs if they do go with a outsourcing strategy for their CPUs.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty
- 0 +

BaronMatrix wrote :

I really don't see how they can make that work as the Fabs would need more than just AMDs business considering that contract foundries are charged less. If you look at Chartered, AMD could get 100K chips from them for (TC) less than making them themselves.

How could the fab survive on that? And then you have to assume that they would just be trading Intel for TSMC as a competitor if they do try to make chips for others.

Moving fab of Radeon or chipsets would also be difficult as GPUs are larger and have less margin while chipsets are even worse (in terms of margins).




Actually, AMD/ATI already outsources fabbing of the Radeon HD family of GPU's to TSMC, they have for quite a while. In fact I think they also outsource Chipset manufacturing to TSMC as well. Which is why the 4850/4870 and most of the current AMD chipsets use TSMC's 55nm node. Otherwise, AMD only really has one operational fab, and thats fab36. Fab30/38 is still in the refitting process.

The main reason for the split would be to get AMD's operational costs down. But Dirks said it's not really necessary for them to drop the fabs to do that.

------------------------------ AMD Phenom X4 9850 Black Edition, ZeroTherm Nirvana 120 Premium CPU Cooler, MSI K9a2 Platinum bios 1.1b3 or P.0J, 4GB (2x2) Mushkin DDR2 1066 (pc8500) 5-5-5-15 2.05v RAM, Sapphire Toxic HD3870, Raidmax RX-700SS PSU, Seagate Barracuda 7200.10 320gb SATA2 X
Reply to Mathos

Good points.

BaronMatrix wrote :

By then AMD should have at least HighK if not preparing for metal gates. It was said earlier this year that they would implement HighK first (D0) and then metal gates (E0). Shanghai is C0.



Considering that you *can't* do highK *without* a metal gate (or a gate with some other different fermi energy), you've heard wrong.

Reply to ryman554

^I didn't even notice that.

I wounder what metal IBM plans to use if not Hafnium. I like that word. Hafnium. It sounds funny.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty

I thought only good companies slit?

------------------------------ AMD X2 6000+ @3.2Ghz | Asus crosshair | 4gb corsair xms ddr2 800 | BFG 8800GTS (g92)x2 SLI 805/1080 | SoundMax HD | (160gb)x2 sata in raid 0 | 500 gb sata | Lian-Li PC-6070 | antec 850W PSU | thermaltake water cooling | Vista 64bit | LG 24" | Logitech 5.1
Reply to ryanthesav

jimmysmitty wrote :


I like that word. Hafnium. It sounds funny.

 


Not as good as Unobtainium :D

Message quoted 3 times
Message edited by Amiga500 on 08-13-2008 at 11:51:40 PM
Reply to Amiga500

ryman554 wrote :


Considering that you *can't* do highK *without* a metal gate (or a gate with some other different fermi energy), you've heard wrong.



Further demonstrates how little he knows in fabrication technology.

------------------------------ Intel will not take the top spot, or probably the top 3 spot back for the forseeable future. Not even with 32nm and more cores will intel be able to beat Jaguar. - JennyH the AMDiot, Nov 2009
Reply to yomamafor1

Amiga500 wrote :

Not as good as Unobtainium :D



Nice. I like that one too. Metals have funny names.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty
- 0 +

ryman554 wrote :

Good points.



Considering that you *can't* do highK *without* a metal gate (or a gate with some other different fermi energy), you've heard wrong.



It's a question of perception. Most consider anything with a higher resistance than standard gate dielectric to be high K.
Most fabs are using nitrogen to effect this at present.
Nitrided oxides are not as effective, but they fit the bill.
AMD (and Baron) is not really wrong, just slightly.

Reply to endyen
- -1 +

jimmysmitty wrote :

Nice. I like that one too. Metals have funny names.



No no no, you missed it.


Unobtain(able)ium...



Slang for the impossible dream material.

Reply to Amiga500

^I know.

I noticed that threads that are important or about a companies future tend to not get any attention and ones that tend to end in flame wars get a lot of attention.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty

Well the craic is all about fusion.


AMD and TSMC need to find a common manufacturing process for fusion. This "asset smart" strategy will undoubtedly play into that.

Reply to Amiga500
- -1 +

ryanthesav wrote :

I thought only good companies slit?



And Elves, they split too....

"Sargeant, bring me another Elf, this one split!"

Amiga500 wrote :

Not as good as Unobtainium :D



And the other super metal nonamium.

------------------------------ AMD Phenom X4 9850 Black Edition, ZeroTherm Nirvana 120 Premium CPU Cooler, MSI K9a2 Platinum bios 1.1b3 or P.0J, 4GB (2x2) Mushkin DDR2 1066 (pc8500) 5-5-5-15 2.05v RAM, Sapphire Toxic HD3870, Raidmax RX-700SS PSU, Seagate Barracuda 7200.10 320gb SATA2 X
Reply to Mathos
- 0 +

IBM's process runs faster SOI ... much faster.

No Hafnium high K gate tech there ... is there?

Please elaborate on that comparison - just asking a question because I don't know...

------------------------------ Great spirits have always encountered violent opposition from mediocre minds

 

Reply to reynod

ryman554 wrote :

Good points.



Considering that you *can't* do highK *without* a metal gate (or a gate with some other different fermi energy), you've heard wrong.




That was the word. I'm not a CPU engr so I couldn't say. I do remember that AMD wil do gate first with metal so it's possible they can do HiK without. Again, I'm speculating based on AMD statements. I'll see if I can find the link.

Reply to BaronMatrix

yomamafor1 wrote :

Further demonstrates how little he knows in fabrication technology.





That was a really trollish thing to say. I found the link. It was our old friend Fudzilla.

http://www.fudzilla.com/index.php? [...] 6&Itemid=1

And here's another from EETimes which mentions High K dielectrics separate from High K metal gates. The assumption therefore is that the dielectrics come first and the metal gates second. It was even hinted that they may forego them until 32nm.

http://www.eetimes.com/news/semi/s [...] =204801525

Reply to BaronMatrix

Amiga500 wrote :

Well the craic is all about fusion.


AMD and TSMC need to find a common manufacturing process for fusion. This "asset smart" strategy will undoubtedly play into that.





TSMC has already qualified their SOI process for 40nm and their SiGe process is also ready for 40nm. The thing about the first Fusion will be the MCM. Later on the GPU will just become the North Bridge, but before then Radeon has to move to SOI. Of course only the APUs will get SOI and I would assume the rest would stay bulk.

The beauty of AMDs arch is that all thats needed is an HT3 tunnel to PCIe. I don't know if it would be efficient to forego the tunnel when they're on the same chip though.

Reply to BaronMatrix

^For Fusion why don't they just link the GPU directly to the HT3 links? Wouldn't that be faster and more efficient for that type of CPU?

Linking the GPU to PCIe then the HT3 to PCIe would cause a bit higher latency in my mind because the PCIe is still on the NB even for AMD chips.

------------------------------ http://www.steamcalculator.com/76561197970703804/camo_sig.png
Reply to jimmysmitty

Amiga500 wrote :

Not as good as Unobtainium :D



This would make a kewl name for a character in WOW :D

------------------------------ Teamwork involves embracing all ideas whether good or bad
Reply to Luminaris

jimmysmitty wrote :

^For Fusion why don't they just link the GPU directly to the HT3 links? Wouldn't that be faster and more efficient for that type of CPU?

Linking the GPU to PCIe then the HT3 to PCIe would cause a bit higher latency in my mind because the PCIe is still on the NB even for AMD chips.




You can't directly connect a GPU to HT3 as far as I know as the I\O is in PCIe packets. AMD had PCI-X tunnels in their server chipsets, so it should be as simple as substituting PCIe at the GPU side of the tunnel. Like I said, when they are not on MCM the plan, supposedly, is to use the GPU as the North Bridge and have PCIe on die. You still need a link to it though. That is unless they can adapt the XBar so that it can communicate directly with the GPU core. That would save space maybe but would require lots of active logic as the GPU is not X86.

Reply to BaronMatrix

Thread : Is AMD going to split?

MHO ....

If there is anything more than 'smoke' to this I don't think a 'split' would be functional or 'technically' operational in nature.

I will guess that there is little to this other than a legal reasoning to the establishment of a wholly-owned operating subsidiary of the 'parent company' AMD.

One thing I'm certain of ... they have paid a bunch of lawyers a boat load of money to set up something like this for regulatory, legal and tax purposes ...

Reply to wisecracker

BaronMatrix wrote :

That was the word. I'm not a CPU engr so I couldn't say. I do remember that AMD wil do gate first with metal so it's possible they can do HiK without. Again, I'm speculating based on AMD statements. I'll see if I can find the link.



You are misinterpreting those links.

The links say "AMD 45nm hi-k" -- and you take that to mean without MG.

But, the links *also* go on to say:

Quote :


Intel rolled out 45-nm high-k chips in November. Pellerin said AMD is focusing more on customer applications for its high-k chips rather than the process technology race.



These guys are using hi-k as the *shorthand* for hik/mg. Unless you can point me to something hich "the word" *specifically* says "hi-k without MG", you're taking the wrong conclusions from the articles.

Quote :


IBM announced a "high-k gate first" process for 32-nm chips earlier this week as part of its extension of a high-k/metal gate development effort with partners AMD, Chartered, Freescale, Infineon and Samsung.



AMD/IBM are jointly working on this,with the rest of the consotrium -- it's the nature of the beast -- if one company develops something, they *all* get it (and IBM gets their cut. They've got a sweet deal as the ultimate middle management). This (or something like it) will be AMD's 45nm offering -- a pilot of what they can do on 32nm. To do otherwise would be a) wasteful, and b) against the rules of the consortium.

Reply to ryman554
- 0 +

You know, during the Clinton era, the DoD was downsizing. Downsizing is a such a negative phrase though.......sounds wrong, like your losing something you may not want to lose. Makes voters unhappy, which makes politicians unhappy, which makes DoD unhappy. Interestingly, DoD changed its phrase to 'Right-sizing'. Sounds much better, like your keeping everything you need, but not that which you dont. In reality, nothing about the 're-sizing' changed but the word used to describe it.

"Asset Lite" to "Asset Smart". Hmmmmm......

------------------------------ http://i249.photobucket.com/albums/gg233/turpit/SIG2A.jpg
Reply to turpit

BaronMatrix wrote :

That was a really trollish thing to say. I found the link. It was our old friend Fudzilla.

http://www.fudzilla.com/index.php? [...] 6&Itemid=1

And here's another from EETimes which mentions High K dielectrics separate from High K metal gates. The assumption therefore is that the dielectrics come first and the metal gates second. It was even hinted that they may forego them until 32nm.

http://www.eetimes.com/news/semi/s [...] =204801525

R

ryman pretty much summed it up. If I remembered correctly, you have to implement both Hi-K and Metal Gates at the same time due to depository reasons. Intel actually wrote a brilliant article on this. I need to ask Jack to find it again for me.

http://www.semiconductor.net/article/CA6494256.html
This is IBM's gate first approach:
http://a330.g.akamai.net/7/330/2540/20080720152529/www.semiconductor.net/articles/images/SI/20071101/six0711_high4a.gif

This is the "gate last" approach:
http://a330.g.akamai.net/7/330/2540/20080720152529/www.semiconductor.net/articles/images/SI/20071101/six0711_high3a.gif

Again, this further demonstrates how little you know in fabrication technology.

------------------------------ Intel will not take the top spot, or probably the top 3 spot back for the forseeable future. Not even with 32nm and more cores will intel be able to beat Jaguar. - JennyH the AMDiot, Nov 2009
Reply to yomamafor1
- -1 +

Thanks for the info ... that is an area I know little about and it is very interesting.

Why didn't you just get Jack to post it ... LOL.

Jack lost his tongue ... er fingers??

------------------------------ Great spirits have always encountered violent opposition from mediocre minds

 

Reply to reynod

I'm not certain but I think what BM is referring to is an incremental step for AMD between now and HkMG (which could be introduced in a second spin at 45nm).

IIRC (which is always a stretch :P) it involved 'low-K dielectrics' in an immersion lithography process which simplified fabrication. I think it was identified as less expensive in their 'process' compared to HkMG at 45nm - which they could have done however 'cost-ineffective' it may have been.

The end result of whatever it is: lower cost, higher margins and 15-20% better performance - something AMD can certainly use ...

Reply to wisecracker

yomamafor1 wrote :

R

ryman pretty much summed it up. If I remembered correctly, you have to implement both Hi-K and Metal Gates at the same time due to depository reasons. Intel actually wrote a brilliant article on this. I need to ask Jack to find it again for me.



Again, this further demonstrates how little you know in fabrication technology.





I SAID I WASN'T A CPU ENGR!

The articles I linked to implied that High K dielectrics do not depend on metal gates, but perhaps metal gates need High K dielectrics. Even the pic you posted showed the dielectric as a separate layer. You should really get a life. Being annoying isn't a skill.

BTW, wasn't that name supposed to be stricken from the forum?

Reply to BaronMatrix

No. As ryman pointed out, the article used "Hi-k" as a shorthand for "HK/MG". Anyone who spends a little time in researching such technology (which, apparently you didn't) knows that Hi-K cannot be implemented without the use of metal gate. So either you misintepreted the article, or you simply do not bother in researching it before posting.

Quote :

You should really get a life. Being annoying isn't a skill.


Describing yourself?

------------------------------ Intel will not take the top spot, or probably the top 3 spot back for the forseeable future. Not even with 32nm and more cores will intel be able to beat Jaguar. - JennyH the AMDiot, Nov 2009
Reply to yomamafor1
- 2 +

BaronMatrix wrote :

I SAID I WASN'T A CPU ENGR!

The articles I linked to implied that High K dielectrics do not depend on metal gates, but perhaps metal gates need High K dielectrics. Even the pic you posted showed the dielectric as a separate layer. You should really get a life. Being annoying isn't a skill.

BTW, wasn't that name supposed to be stricken from the forum?



Is passing incorrect/incomplete/out of context information a skill?

------------------------------ http://i249.photobucket.com/albums/gg233/turpit/SIG2A.jpg
Reply to turpit

wisecracker wrote :

I'm not certain but I think what BM is referring to is an incremental step for AMD between now and HkMG (which could be introduced in a second spin at 45nm).

IIRC (which is always a stretch :P) it involved 'low-K dielectrics' in an immersion lithography process which simplified fabrication. I think it was identified as less expensive in their 'process' compared to HkMG at 45nm - which they could have done however 'cost-ineffective' it may have been.

The end result of whatever it is: lower cost, higher margins and 15-20% better performance - something AMD can certainly use ...



Except that the lo-k dielectrics are for the back-end (metal lines) *not* the front end (transistors).

That, in and of itself, won't get you any more fundamental switching speed -- only a lower RC overall. I can't tell you what that does to performance -- it's certainly better, all other things being equal (and, sadly, yield/reliability tends *not* to be) -- I'm not an EE.

And I *know* that AMD is claiming a respin on 45nm to implement HKMG. From a manufacturing standpoint, that's crazy. That means, instead of developing one process over two years, you're doing one every year. For the next three years (45nm in Q408 being the first) That's nuts. If you're going to do that, you might as well just make the move to 32nm while you're at it. Changing the front end (even if "seamless" ) changes *everything*. Now *demonstrating* HKMG on 45nm -- *that's* smart. Gets a lot of the kinks out before 32nm (the real deal) comes along. Ramping to HVM? Color me skeptical. I'll be glad to eat crow over this.

Reply to ryman554

If you're not an EE, I don't know what EE is... :p


Message edited by yomamafor1 on 08-16-2008 at 03:11:34 AM
------------------------------ Intel will not take the top spot, or probably the top 3 spot back for the forseeable future. Not even with 32nm and more cores will intel be able to beat Jaguar. - JennyH the AMDiot, Nov 2009
Reply to yomamafor1
Tom's Hardware > Forum > CPU & Components > CPUs > Is AMD going to split?
Go to:

There are 1197 identified and unidentified users. To see the list of identified users, Click here.

Please mind

You are about to answer a thread that has been inactive for more than 6 months.
If you still wish to proceed, please ensure that your posting is original and does not duplicate or overlap any prior responses to this thread.

Add a reply Cancel
Sponsored links
  • Ask the community now
  • Publish
Ad
They won a badge
Join us in greeting them