When I say incompetent I think I'm already implying human factors. AMD management in general is very incompetent. That they tried to portrait something like Bulldozer as a server processor is just the icing on the cake. They didn't hide the complete truth, they screwed up big time in the marketing message.
It could've been the other way round. They had the plan for another server processor from the beginning. That's as business is expected to happen if it is full of MBAs. The R&D teams and engineers have to deliver. But with such complex things as CPUs where you've still to find the solutions, which are already scheduled and tied to milestones, there could always be delays, buggy features, wrong assumptions (heuristics), imprecise simulations, etc. This is the human factor at work at every level.
The decline was not because AMD didn't do much, but because they failed in everything they tried to do after K10. They had to flush the entire R&D pipeline because of the Bulldozer fiasco.
When did they flush the entire pipeline? They left out finishing some projects like Krishna, Wichita, Skybridge, and others, which was either related to GF processes, market situations or changed plans. The last big µarch thing being cancelled which I know of is David Christie's (and even Jim Keller's) original K8 design. Nothing is really missing in the construction line. Ok, the only thing I might miss from the original µarch roadmap is "greater parallelism" of SR, but this is interpretation (in the end +0.1% is "greater"). Other topics like ASF have been developed, but didn't get a go - possibly due to discontinueing the server line. Otherwise they'd have announced it, since ASF is nothing else than TSX.
And regarding servers, Johan pretty much summed it up:
AMD promised us (in 2009/2010) that the Opteron 6200 would be significantly faster than the 6100: "unprecedented server performance gains". That is somewhat the case if you recompile your software with the latest and greatest optimized compiler as AMD's own SPEC CINT (+19%), CFP 2006 (+11%) and Linpack benchmarks (+32%) show.
One of the real advantages of a new processor architecture (prime examples where the K7 and K8) is if it performs well in older software too, without requiring a recompile.
http://www.anandtech.com/show/5058/amds-opteron-interlagos-6200/14
Chuck Moore proposed throughput (more threads) being the solution. And he (or someone else) decided to take the risk of not getting enough compiler support and lower performance on existing code and code compiled for other archs. This is a decision and didn't play out. But that's economic life in a perfection seeking world. Nobody has a crystal ball. And only bigger companies have enough power to ensure the following of the software crowd.
LOL. Are you serious?
AMD's decline could be more attributed to him joining AMD, than leaving it.
He made an idiot out of himself on just about every computer forum, insisting Bulldozer wouldn't suffer IPC loss.
Hey, you detected my sarcasm.
If you looked closely, I had both positive and negative points in this list. So the kind of effect of JF's work is up to you. I think the most likely thing is that we take such forums as too important as nearly noone among our engineers knows them (sometimes unfortunately, sometimes fortunately). And we're doing ADAS and more future stuff.
Facebook
Twitter