Yup, Semiaccurate reported on this story a week ago. Its probably a good move, given that the slips in GF's 28nm process mean that the two products would only have been produced for 6 months or so.
SemiAccurate got the story wrong and blames the issue on GF pushing back their SHP process. They don't mention the move to TSMC and they claim there will be a follow-up in months.
Auxillary Power Unit: small electric thrusters near the pivot point of a ship used in emergencies or certain close quarters situations, such as docking.
Ambulatory Procedure Unit: where surgeons do same-day procedures.
Navy, medicine, computers: inside my head, acronym reuse is a real problem.
Are these Bulldozers really as bad as I keep reading about?
A friend of mine who is very knowledgeable when it comes to hardware insists that the issues are being overblown, and that if you get the correct configuration of hardware (ram/mobo/etc) along with the right overclocking setup, these procs are just as good, as well as more future-proof. He says most benchmarking tests are more single-threaded examples of load, which the Bulldozer obviously performs worse with, despite this being a more realistic representation of the kind of load you'd find in your average desktop, especially when it comes to gaming.
As you note, for desktop/gaming purposes it's pretty obvious that single- or few-threaded performance is still king, and I see no reason to expect that to change in the foreseeable future. And Bulldozer is really, really, really bad at it. "But you can overclock it!" is a silly argument; you can get an i7 up to 5GHz or something equally unnecessary and it'll blow away whatever you can get that Bulldozer silicon up to. The bigger problem is in the hardware design, which is intensely over-shared and results in hardware-level blocking conditions, as evidenced by the various reviews out there...and overclocking doesn't help that.
I'd consider Magny-Cours for some types of server workloads, though I'd probably go with Sandy Bridge (and definitely would for a desktop). I wouldn't buy Bulldozer for anything.
I'm looking into a new rig, and he's completely sold on the design. I can imagine a world in, lets say, 2013-2014 where the desktop becomes a more multithreaded environment, but that just isn't where were at today, and thats just my guess. He's convinced the overclocking aspect makes all the difference, and Intels don't OC as well, but thats not what i'm reading (nor what you're saying).
I do like the conceptual architectural changes made with the bulldozer, but current, and forthcoming, software just doesn't seem like it will make use of it. It definitely seems like more of a server-minded approach to an architecture.
I think you're being optimistic. We're going to see all our desktop applications become pervasively multi-threaded in two years?
And Bulldozer is going to be better at this than Sandy Bridge, which is good at both single- and multi-threaded loads?
Ehhh. Not likely. The design isn't even that good or interesting; as I mentioned before, it's overly reliant on shared components that aren't conducive to the sort of magical perf improvements that "but you can overclock!" would require.
Taken over mobile market? That's quite an exaggeration. Yes, Llanos were relatively competitive, but large OEMs and markets do not change that quickly. And Intel and Sandy Bridge got lots of good press too.
Quick googling shows that Intel's market share in laptops is still 82%, so while AMD has certainly gained some traction, it is still far from "taken over" any markets.
I feel like this is a great time to make the argument that AMD chased short term profitability ahead of long term profitability. The pushed the fabs and debt off their books, and in the process handicapped themselves for the future.
They really had no control over the engineering stack necessary to push innovation and keep pace with Intel. It's especially difficult if you cannot have your own engineers intimately involved in manufacturing to go back to design and make the necessary changes to keep pace with your competitors who can do this. Has fab-less worked before? Sure, but that's a different ball game than the market AMD is in, often shipping much lower volumes and in lower power devices with smaller die sizes.
And obviously, this on top of other management blunders at AMD.
http://semiaccurate.com/2011/11/15/exclusive-amd-kills-wichi...