Originally Posted by MrJava
I honestly don't you can completely rework a CPU design that quickly.
Originally Posted by sdlvx
The term I like to use for AMD's situation in regards to Bulldozer and derivatives is "abundant low hanging fruit"
Friendly reminder it was announced August 2012 that Jim Keller returned to AMD. Kind of convenient timing regarding time timing of cancellation of bdver3a and bdver3b.
Also, digging through old articles regarding Keller's return, I notice a lot of comments talking about how AMD needs to go back to drawing everything by hand. And now we're talking about how AMD is going to basically get a free die shrink by re-doing the part that was done by hand with automated tools.
That would be nearly 2 years of development time for a revision of the chip small enough to not warrant a new code name. We also don't know if bdver3b was in development before that. However i'm not implying a complete re-work. As I said there's "abundant low hanging fruit" so it should be possible to get very good gains without massive re-working of the chip.
I guess what I'm trying to say is that bdver3a might have had some easy to fix problems that would net a lot of performance that someone like Keller would pick up on and others would miss, and AMD may have calculated it would be best to can bdver3a in some markets, develop bdver3b.
I do realize that 2 years to retail is not enough time to make drastic changes, but I would think of it more along the lines of a better stepping of a chip as opposed to a new microarchitecture.
It just means things will be interesting. Perhaps AMD will kind of mimic Intel with the "enthusiast" class being delayed from mainstream, but AMD might up the ante a little bit by making performance oriented changes to the architecture. It would make a lot of sense as Intel seems to be wanting to cram their laptop architecture into server chips and performance parts, and it's leaving a lot of people wanting more. That and it's not providing lucrative reasons for people to upgrade, from say SB overclocked to Haswell overclocked.
If AMD could deliver an enthusiast platform with 15% or so increase in performance per core a year while adding more cores or whatever, they'd give a lot of enthusiasts who want to buy new toys a reason to buy new toys, instead of buying "herp derp shake and bake raise dat multi i'm so enthusiast!!!" update that's 5% faster, you'd get something good.
Of course I'm assuming AMD is still going to care about their "enthusiast" market and they might give up completely.
Also, I always call it "enthusiast" because it's just a marketing term between each company and according to Intel, everyone who doesn't use LGA2011 is "mainstream". I don't buy that, iPad or Tablet or whatever is mainstream.
Regardless, it sucks to think about it but June 2012 is going to be two years by the time we'll see bdver3b if it exists.