NovusBogus wrote:Wishful thinking on Anand's part. I could see them reviving K10 but the shrinking PC market doesn't have enough money to justify Intel or AMD developing a new CPU architecture for a very long time.
Going back to the 2nd post in the thread, this is totally wrong. Is the assumption above that AMD is going to develop a new architecture after the next one or two iterations of the old one? If so, that's false. Both AMD and Intel and Nvidia and any other processor design house work on these things for years before anyone hears anything about them - and my guess is that the timetable is longer for CPUs than for GPUs. AMD likely had a rough roadmap for their next architecture before the first Bulldozer hit the shelves. Then, I'd wager that roadmap saw some changes after Dirk was shown the door, based upon the weaknesses of the Bulldozer architecture. Since those early days, I imagine that roadmap has been continuously refined and is no longer a roadmap, rather it is a full technical specification at this point. Much of the next architecture is already designed at this point, major features are implemented and "frozen", as the engineers continually close in on the smaller and smaller details.
The PC market is still huge. So is the laptop market and the server market. These markets have always used variations of the same architecture.
Intel seems to have reached a point where it has a fundamental architecture that it may stick with for a very long time and will just continually refine it unless and until some breakthrough is discovered.
AMD's fundamental architecture has some big issues: power consumption, max attainable performance, even performance, etc., etc. AMD will bring out another X86/X64 architecture or it will eventually have to exit the X86 market, because Intel won't stop until AMD is no longer in the market.
The words from the Anandtech article don't sound plucked out of the nearest a-hole. They're also not extremely specific. The article says that one more major revision of Bulldozer will launch in 2015 and that it will be the last, but that does not present a date-certain for the launch of the next architecture. It sounds like the earliest we could expect that would be 2016, which would be a full five years after the launch of the FX 8150. But it could stretch to 2017, too - AMD is good at stretching like that!
The big question is whether AMD's next architecture will save anything from the Bulldozer era. There has been a lot of APU work done. It seems strange to imagine "back-porting" all that APU work to the Deneb/Thuban architecture, but that's not to say it is impossible. What is interesting is that I always saw Bulldozer as an attempt to deliver hyperthreading on steroids, so if AMD ditches the module approach, will it implement some form of hyperthreading? From a marketing standpoint, it would be a big disadvantage not to have some kind of counter to Intel's hyperthreading.
Finally, it should go without saying, that's all just a bunch of guesswork and speculation. My main point is that I think the comments in Anandtech's article were not speculation, though.