Will AMD Buy ATI?
No. Why? They have no money, and any money they do have is going to have to go into their next fab. It’s hard to beg for subsidies from German governments to redo fabs (see below) when you’re spending your money buying other big companies.
Nor is their share price isn’t so high that they could issue relatively few additional shares to buy ATI. Indeed, AMD’s stock price has dropped 25% in the last three months, which means they’d have to issue a lot more shares to buy ATI.
Nor is there any pressing need to buy ATI? Intel does video, too? Intel’s been doing video an awfully long time. They’ve been doing motherboards, too, does that mean AMD has to buy Asus all of a sudden?
The only way AMD might seriously consider buying ATI is if Intel decided that they were going to make a big move on the video front: a superintegrated video system or some sort of CPU-GPU hybrid and bought one of the big graphics card companies. There’s no signs Intel wants to do anything like that, though there might be on AMD’s end (see below).
Even if this came up out of the blue, I suspect AMD would be quite reluctant to buy a video card company, simply because it would be a considerable distraction at a time when they need to be very, very focused on their core business.
What About This New Fab?
Not new, refurbished. They’re going to reequip good old Fab 30 and give it a new name. I suppose that will cut the cost somewhat, and retain the current jobs at Fab 30 (when AMD got the subsidies for Fab 30, they agreed that Fab 30 would provide X number of jobs until 2008).
What hasn’t been mentioned so far is how much the state and national German governments are ponying up for this; the political environment is probably a little less friendly than it was when Fab 30/36 was financed.
Presumably, this will be a 45nm fab.
Sure looks like a warmed over Opteron platform aimed at those with more money than sense to me.
You can read the story here but it boils down to:
Accelerator here, accelerator there, accelerator, accelerator anywhere. Literally, anywhere from a PCIe card to the chipset to a CPU package to a CPU core.
Now this could well be a great thing for specialized computers, but, outside of a PCI-E accelerator or maybe a co-processor socket, this is not going to come to a desktop near you. Why? Just how are you going to mass-produce these things?
If twenty companies come up with something that can fit inside a chipset, or within a CPU package, or a CPU core, what’s AMD going to do? Design twenty different CPUs, or chipsets, and run twenty different CPU batches? Yes, they could do that, but do you realize how expensive that is? Sure, if paying thousands of dollars for a specialized chip is no big deal for your computing project, that’s not a problem, but only there.
So, for the desktop, I’m afraid any “accelerator” technology is going to be 95% PCI-E cards and 5% CPU socket, but even that could prove to be a very good thing.
Why? It could create a need for non-video big PCI-E (or maybe something new) slots. We’ve pointed out that technologies like the iRAM were grossly bottlenecked by the speed of the PCI bus, and noted that a 1X PCI-E slot wasn’t exactly a solution.
Accelerator cards will probably need at least a multilane PCI-E slot to perform well, too. What might be even better down the road would be if AMD could somehow take a Hypertransport 3 link and bring that to expansion cards augmenting/replacing PCI-E.
That seems to
be the direction they want to go: linking everything up with hyperfast connections. A big part of the advantage Opterons have over Xeons is Hypertransport, and it looks like AMD wants to beat Intel over the head with that advantage any which way they can and solidify the market share they’ve gained the last year.
Eventually, though, that idea will move towards the desktop, though probably not until towards the end of the decade.