Anandtech: AMD-ATI Merger in the Works?

JoshMST said:
If AMD were to merge with ATI, NVIDIA would not intially drop out of the AMD chipset market, but it would cozy up closer to Intel. Not to mention that ATI would then probably be allocated Fab space, and advanced process nodes, well before NVIDIA could do so with TSMC (which looks to be about 6 to 9 months behind AMD when it comes to process node changes). I think that such a merger would be bad for the industry.

Why and how would it be bad for the industry?

If Intel snagged ATI or NV that could be bad because of Intel's market position and ability to use their market share to leverage a GPU maker. As Intel already is the #1 GPU provider in volume snagging ATI or NV would solidify that market position to a point that there would be no competition.

AMD on the other hand does not make GPUs, nor do they dominate the market like Intel. AMD has less than 20% of the total PC market, so while AMD could leverage their connections with OEMs and at retail, it would be much smaller and still allow room for Intel and NV to compete.

From a consumer perspective if AMD bought ATI or NV it would give the GPU maker quicker transitions in process changes, access to more engineers, and possibly more influence on the direction of the industry. This could mean some products with tighter integration, but realistically what we could see is small, faster, and cooler GPUs 6-9 months sooner than we are now. Probably not a good thing for NV if AMD bought ATI, but on the other side it means ATI could possibly provide denser or smaller chips which does help consumers.

I doubt this will happen, but it is interesting. Based on AMDs relationship with NV I thought that would make more sense. On the one hand it would be nice to see stability in the industry and to ensure both NV and ATI are around for a while, but AMD worries me. Conroe looks very tough (even without integrated memory controller) and the AMD roadmap for ZRAM and Clearspeed look to be more down the road. This could hurt ATI as much as it helps depending on the direction AMD drives ATI.

Down the road this could be interesting. MS went with IBM this last time around for consoles due to cost control and IP issues. If ATI is snatched up MS may lose some of those options with AMD. If MS wants a custom GPU and CPU and ATI is aligned with AMD that could create some interesting scenarios, but it also could be more expensive and lose control, something MS wouldn't want. Of course this could also lead to some interesting GPU-CPU integration as well. Who knows...

Next week it will be Intel buying ATI... or NV.
 
The the increased chip volume of having ATI products produced on AMD fabs is why this makes sense to me.

Could be the next Xbox = AMD CPU and ATI GPU both made at AMD fabs
 
Brimstone said:
The the increased chip volume of having ATI products produced on AMD fabs is why this makes sense to me.
Currently AMD can't even sustain production of enough of their own chips in their own fabs, so that reasoning doesn't seem persuasive to me.
 
Acert93 said:
I doubt this will happen, but it is interesting. Based on AMDs relationship with NV I thought that would make more sense. On the one hand it would be nice to see stability in the industry and to ensure both NV and ATI are around for a while, but AMD worries me. Conroe looks very tough (even without integrated memory controller) and the AMD roadmap for ZRAM and Clearspeed look to be more down the road. This could hurt ATI as much as it helps depending on the direction AMD drives ATI.

Well, that might be the interesting thing to AMD. . .arguably they are at a peak right now, the biggest in company history, with a valley in their core business staring them right in the face. What better time to diversify their revenue stream *and* secure some assets that help the core business as well.

But are they really wanting to pay cash for ATI with that aforementioned valley staring them in the face? That cash is their blankie right now to get thru that. Would ATI really be interested in a stock swap deal with same AMD valley staring them in the face?

Given NV's focus on margin-rich markets, I could even imagine AMD positioning ATI's chipset business as a low-end high-volume "floor" for AMD's markets, with NV keeping the high-end value-add mobo business.

Except. . .then there's CrossFire, and that's a high-end solution and requires high-end-ishy mobos.
 
Come to think of it, a tangential datum. . . would ATI have signed-on a High Road for PR activities if they were actively entertaining offers from much bigger suitors?
 
geo said:
Come to think of it, a tangential datum. . . would ATI have signed-on a High Road for PR activities if they were actively entertaining offers from much bigger suitors?

They're publically traded - they aren't necessarily entertaining offers. AMD may just be shopping.
 
Zaphod said:
Currently AMD can't even sustain production of enough of their own chips in their own fabs, so that reasoning doesn't seem persuasive to me.

That's the short term outlook.

ATI would help insure volume levels (i.e. lowering financial risk) as AMD invests money to increases production capacity (i.e. building new FABS).
 
Whichever company ends up with Intel is better off.

I think ATI is struggling and, teaming up with a struggling CPU maker doesn't make sense.

AMD is in a no win position against Intel. Even when they win for a long period of time as with A64, it gains them little share, and sooner or later Intel's superior finances and R&D budget overwhelm them anyway (Conroe).

If AMD and ATI merge, and on the other hand Intel buys Nvidia, the ATi-AMD combo is in danger of going out of business eventually imo..
 
You do realize that AMD has been going against Intel for a very long time now, if they were going to go out of business it would have been a long time ago.

AMD is in the highest time of its life, its out selling Intel right now, and Intel has been playing catch up since day one of the A64s life. AMD is one step ahead right now, IMO, they will have a bit of harsh period with Conroe, but they'll get over it, they've done it before and they'll do it again.

ATi struggling? They're producing the better chips performance wise, with more features, and better IQ. This is struggling? ATi is also grabinb gobs of the chipset market, even if its low margin its lots of it, they're making money. They're inside the Xbox 360, they're inside the Nintendo Wii.

I dont see you're arguement..... at all.
 
  • Like
Reactions: Geo
I think a market consolidation of that magnitude brings up a lot of negatives. Especially when considering the two markets that the one company would address are currently really only occupied by two companies (CPU's are Intel and AMD, and graphics are NV and ATI obviously). So, AMD and ATI merge, and we see all kinds of shakeups in the chipset market, not to mention some very interesting balances between R&D resources, not to mention fab space.

Right now AMD is finally catching up with backorders due to Fab 36 putting out 90 nm parts, but that fab is also being turned over to 65 nm production as we speak. Once that fab is going full bore, it appears as though AMD will take production space offline in Fab 30 and upconvert it to Fab 38, which involves construction, taking apart the clean room, etc. I don't see that AMD's production capacity will really be much more than it is right now, especially as AMD will create Athlon's with greater die area to keep up with Intel. Even though they will gain space by going 65 nm on 300 mm wafers, K8L cores will be comprised of many more transistors than current Rev. F products, and we will see die sizes of 150 mm square sooner rather than later, even on 65 nm. So where will the extra fab space be for AMD to create parts for ATI? As Peddie mentioned, AMD currently is producing parts with approximately a 60% margin, while ATI has gross margins on their products that are reaching 30%. Would it be in AMD's best interest to give ATI fab space for lower margin parts?

My opinion obviously is that both companies are stronger engaged in partnerships than they are as merged corporations. If AMD were to buy ATI, then they would really alienate NVIDIA, who has been one of AMD's strongest partners through the years.
 
Neither one gains anything that supports its core business. ATI does not have a great reputation for AMD chipsets. AMD doesn't need to worry about video cards for any reason. The only point of overlap I could possibly think of besides core logic is in the consumer space--all-in-one AMD/ATI products. Beyond that? Nothing. ATI's always been fabless, and I don't think AMD has the spare production required to take over everything from TSMC/UMC.

Intel and ATI make a *lot* more sense than AMD/ATI. But even that's silly, so...
 
If this came to pass, I wonder what would happen to the ATI - Intel cross licensing deal going forward.
 
Skrying said:
AMD is in the highest time of its life, its out selling Intel right now

Not true. Intel still has ~80% of CPU sales. AMD has about 80% of the retail consumer CPU sales, but those are a very small portion of the market (something like 10% of my memory is right).

JoshMST said:
I think a market consolidation of that magnitude brings up a lot of negatives. Especially when considering the two markets that the one company would address are currently really only occupied by two companies (CPU's are Intel and AMD, and graphics are NV and ATI obviously).

But as mentioned before, this is not completely true.

The CPU market is much broader than AMD/Intel. One need only look at the conosle market where all three console CPUs have been designed or co-designed by IBM. Over the next 5 years there will be about 150M IBM designed CPUs going into consumer living rooms. Considering Intel and AMD barely shipped over 200M CPUs last year that should put it into perspective. IBM, Sony, Toshiba, NEC, etc all make contributions to the CPU market, with the first two being fairly active and aggressive players.

On the GPU front Intel -- not ATI or NV -- is the leading supplier of GPUs. One could argue aquiring ATI would help AMD compete with Intel because 1) ATI would bring along quite a few engineers capable of working on MB chipsets and 2) with the increasing role of GPUs in desktop computing (see: Vista) AMD is in danger of being left in the cold if the market shifts.

So while a consolidation move, the move actually is one to move AMD into a profitable market they are incapable of competing with Intel on. One could argue if AMD remains out of the GPU market they could be in a dire long term situation.

With ATI on board they can do two things that could help AMD compete with Intel: Offer, just like Intel, their own integrated graphics with a MB chipset.

The second, and possibly more exciting, is the reality that CPUs don't scale in performance quite as well as GPUs by adding extra cores. At some point it may become very attractive for AMD to go a slightly different route. e.g. Instead of an 8-core CPU, why not 6 cores + 1 GPU -- all on die.

Without AMD neither of these are options, and if GPUs become as important to desktop computing as many project AMD not making a move could actually be devistating.

Right now AMD is finally catching up with backorders due to Fab 36 putting out 90 nm parts

ATI can still use TSMC. In fact, it would probably take a bit of effort to move over to AMDs process so it may not even be a pressing issue.

You mention partnerships, but the question is how long would those last if AMD seriously faltered in performance comapred to Intel? Or what happens when Intel aggressively makes a move in regards to the role of graphics? Intel is frequently in the driver's seat and graphics is an area (and chipsets) where AMD has a serious hole in the portfolio.

I don't necessarily like the move (mainly because I don't think AMD is very stable to begin with) but from a consumer and market competition stance this would make AMD stronger, and possibly ATI if fab space ever opened up.
 
Acert93 said:
The second, and possibly more exciting, is the reality that CPUs don't scale in performance quite as well as GPUs by adding extra cores. At some point it may become very attractive for AMD to go a slightly different route. e.g. Instead of an 8-core CPU, why not 6 cores + 1 GPU -- all on die.
What are you talking about. Compare transistor counts, compare thermal output, compare power requirements, and you'll see it's impossible. Assuming we're just moving into Implausibleland and saying "But improved processes will deal with that," it makes absolutely no sense. Why should a GPU be integrated into a CPU? It works fine in the motherboard. The CPU stays simple. The socket stays simple. You still need outboard components for, you know, display. The memory bandwidth sucks just as much as before, and latency isn't a huge issue. it yields a remarkably more complex chip with functionality that a large number of your customers wouldn't even use, and if you're in the chipset market anyway you can provide it there easily. Why do merger threads always bring out the ideas of "We can merge the entire product lines and form a SUPERPRODUCT!" that end up being total nonsense?

Oh yeah. GPUs ARE INHERENTLY PARALLEL. CPUs AREN'T. GPUs process pixels--you can go to the next one and start working on that while you're rendering the current one. CPUs process instruction streams, which may or may not be parallel--you can't just jump around (okay, you can to a very limited degree) unless specifically instructed. Gyargh! I thought this nonsense was kept to the Console Forum!
 
The cpu boys don't typically do the half-nodes (110nm, 80nm, etc), do they? Are they usually far enough ahead that this doesn't matter? Is AMD going to have 65nm going in volume before TSMC has 80nm going in volume?

I can sorta kinda close one eye and tilt my head and see a case for AMD to like such a deal. I think it's harder to make the case for ATI to like the pluses more than the minuses of such a merger, particularly if it's not an all-cash-at-a-premium to the current ATI stock price deal.
 
The Baron said:
What are you talking about. Compare transistor counts, compare thermal output, compare power requirements, and you'll see it's impossible. Assuming we're just moving into Implausibleland and saying "But improved processes will deal with that," it makes absolutely no sense. Why should a GPU be integrated into a CPU?

You are making a major assumption: That GPU would be top of the line, or even midrange. If you look what is integrated into MB chipsets now that should give you a common sense idea of what they could add to the CPU die at some point. Since you mention the consoles, the fact is the console market has demonstrated that integrating CPUs and GPUs onto a single die is possible after a period of time. Whether it makes sense is more dependant on how beneficial additional x86 cores (will 16 cores give a noticable bump over 8?) are and what the market demands. If the market wants smaller, single chip solutions this could be a possibility.

It works fine in the motherboard. The CPU stays simple. The socket stays simple. You still need outboard components for, you know, display.

That is all fine and good, yet from a financial standpoint you open the door for ATI, NV, and Intel to move in on GPU sales. And integrated product may be a good trade in regards to the added expense/additional return. Of course all your comments overlook the benefit of chip consolidation and what that can mean for new markets like the (super cheap computers).

Why do merger threads always bring out the ideas of "We can merge the entire product lines and form a SUPERPRODUCT!" that end up being total nonsense?

The only one talking about super products is you. Considering ATI's flagship GPU consumes about 2x the die space and produces significantly more heat than AMD's processors I thought it would only be common sense we are not discussing R580 chips here. Maybe something more along the line of the Xpress 200 series which are small and cool.

My point was that CPUs are not inherently parallel, which you ask at some point: When is it more beneficial to add something other than "yet another CPU core" to the die? Is broadening the market appeal of your chip by incorpering more robust features something that can help bolster the bottom line?

Take a look at Intel's roadmap: It clearly outlines asymmetric designs at some point in the future. AMD has likewise began researching incorperating dedicated FPUs. The idea that non-traditional processors will be added to processor dies at some point in the future is inevitable. We know GPUs can be incorperated into the same die, the question is whether it will be beneficial to any markets.

The fact the poorly performing display devices on MB chipsets have dominated the market for years now, I think your assumption that it is impractical is hasty to say the least. Ditto integrated sound.
 
Acert93 said:
My point was that CPUs are not inherently parallel, which you ask at some point: When is it more beneficial to add something other than "yet another CPU core" to the die? Is broadening the market appeal of your chip by incorpering more robust features something that can help bolster the bottom line?

Take a look at Intel's roadmap: It clearly outlines asymmetric designs at some point in the future. AMD has likewise began researching incorperating dedicated FPUs. The idea that non-traditional processors will be added to processor dies at some point in the future is inevitable. We know GPUs can be incorperated into the same die, the question is whether it will be beneficial to any markets.

The fact the poorly performing display devices on MB chipsets have dominated the market for years now, I think your assumption that it is impractical is hasty to say the least. Ditto integrated sound.
You totally, totally misunderstand Intel's roadmap. Completely. Asymmetric processor cores are coming, yes. Hell, they're already here--see Cell. That's the extent of it, though. You're going to see processors with a certain number of general-purpose cores and then a certain number of more special-purpose cores. However, those special-purpose cores are not going to be GPUs. They're most likely to be vector units (check out the recent benchmarks on G70 versus P4 in FFTs) as superscalar architectures are not able to handle vectors effectively at all. But they are not going to be GPUs or sound chips or anything like that. They will be units focused on accelerating specific instruction sets.

General-purpose CPUs are not inherently parallel, but software can be. Look up Amdahl's Law. There is an upper-limit on parallelism in software, although you'll get plenty more benefit from multitasking and the like.

Consolidation also screws up the supply chain. You add a premium to the most fundamental component for something that might not necessarily be used, and if you have chips with an integrated GPU and chips without, you also drastically confuse your supply chain.

It's not better, it's not even worth considering. It's just silly.
 
Back
Top