Bulldozer architecture - so bad it's a conspiracy?

AMD made the same mistake Intel made with the Pentium 4.

That said it's a true 8 core and has nothing to do with hyperthreading. Hyperthreading is a SOFTWARE solution... it's a vertual core. The intel fanboys love to claim hyperthreading is some sort of "full thread" solution... yet it doesn't work like that, and all it is is a schedualing gymic that might result in as much as a 20%-30% boost in performance if the program can make use of it.

The result is each hyperthreaded core is roughly equal to a 1.25 core, in the right circumstances. So a 4 core hyperthreaded i7 is roughly equal to a 5 true core processor (in apps that can use hyperthreading).

AMD's 8 core 4 module system is an 8 physical core system... this isn't hyperthreading scheduling magic. What AMD has done is taken certain resources that a standard core usually has all to itself and shared some of them between 2 cores, creating the module. The idea was with properly timed scheduling you could safely share certain core resources allowing you to pack more cores into a smaller area, and make it more energy efficient to boot.

The problem with bulldozer is 2 fold, first of all, scheduling remains a disaster... the core modules don't work particularly well, and to improve their performance AMD lengthened the pipeline, a design concept intel tried to implement into the Pentium 4 (to generally disastrous results). The theory is sound, and lured intel in nicely 10+ years ago with the promise of a nearly unlimited ghz scaling. Unfortunately this type of gimmick also increases latencies throughout the design which of course then introduces performance bottlenecks independent of clock speed, something intel learned with the prescott... and AMD is now trying to fix with the bulldozer.

So we have a chip with terrible latencies throughout, poor scheduling preventing the cores to function properly and efficiency, a larger nm size because global foundries sucks, and the longer pipeline introducing all sorts of power and heat inefficiencies resulting in a lot worse actual performance then i'm sure the engineering team ever imagined was possible.

The theory is quite clever, the implementation sucks. Its believed that steamroller should fix most of the remaining structural problems with the original bulldozer design, including fixing the core modules so that they work properly. If true it should represent a pretty significant improvement in chip performance. the problem is they've lost so much time screwing around with the problems with bulldozer its doubtful even if they fix all the problems and give it a sizable boost in performance that a single core of the steamroller chip will even equal a 4 year old first generation lga 1156 core.

they're falling so far behind thanks to this mess up it will take a minor miracle to remain relevant much longer in the desktop market.