"Bulldozer will be awesome!" They say.
Bulldozer sucks.
"Piledriver will be awesome!" They say.
Piledriver sucks.
"Steamroller will be awesome!" They say.
...
Do we really need to wait and see?
Firstly most people with a brain said Bulldozer will offer improvements, and potentially be meh elsewhere. Braindead people went "new equals instantly better always therefore bulldozer will wipe the floor with everything", it didn't and those same people went "its crap because it doesn't win super pi, zomg, lame".
It gained what, beyond 10% simply from a windows scheduler update, it gains more than that from proper optimisation. Software completely unoptimised for ANY chip will run like absolute turd vs software completely or even a little bit optimised for it.
Bulldozer is still no where near as bad as people say, there were still 4-5 area's it DID beat sandybridge, and other area's it lost(marginally) to a hex core Phenom 2, boo hoo.
I said as did many others, first chip in a new architecture will be the worst chip you'll see for 5-10 years depending on how long said architecture lasts. It is the least optimised chip in terms of layout, design AND support from software, OS. The second gen both had, an updated scheduler on OS's before its release, more software aware of optimisations for it(but still the vast vast majority run on Intel compilers that do very little to help run on AMD chips at all well), and fixed literally dozens of smaller problems, but it was also WITHIN A YEAR of the previous chip and MAJOR fixes were NEVER going to be in a chip that close. Here's a hint, GPU's take 7-8 months to tape out, so for instance if the 5870 came out, and there needed to be a massive beast of a change in the 6970 it would need to be tested post 5870 release essentially, found, fixed, tested, redesigned in multiple stages, improved, then be ready to go in the final product within 3 months to be ready to tape out for a year later.
CPU's take 18months + to tape out usually, you do not generally get major "fixes" from one iteration to another, think tick tock for Intel. Small changes, bug fixes, minor improvements after a year, after two years, big changes.
This is no different, anyone with any sense who doesn't go "new therefore instantly perfect" knew bulldozer would be a step forward in some area's, compete bettere with Sandy in some area's, and be worse in some area's, and certainly have a problem or two that needs fixing. It was also plain as day that some of these would(and did) get fixed in PIledriver and bigger fixes would be coming in Steamroller.
These things were OBVIOUS to a sensible person 3-4 years ago, because every COMPLETELY NEW architecture has the SAME issue, its always the worst and always has the most to be gained essentially on a long line of upgrades to the architecture. Its also not disimilar to the design process for almost everything made in the world.
Ipad, ipad 2 ipad 3, (okay I don't follow apple so the order or numbers might be off but) simple basic flawed version, followed by version with a number of fixes and small easy improvements, then a version with a completely new chip, vastly more power, loads of the kinks worked out, smoother this and that, streamlined something else, etc, etc.
You don't go into a workshop, design a chip, 3 months later its finish, 3 months after that ready for sale and a year later you can have a completely reworked chip ready, Intel can't do this with billions upon billions at their disposal, ARM chip makers STILL haven't all got A15's out, no a15 quad cores, and theoretically they could have been made the second 28nm was up and running which was well over a year ago. Chip design, even when half the companies use a basically already existing design with minor tweaks, take a ruddy long time.
Only people who were disappointed with Bulldozer were those who can't read benchmarks(it beat Sandy in places the X6 couldn't touch it), and those who for no apparent reason expected the second coming of jesus.
It was a tiny bit slower than I expected on launch, but with scheduler updates, and when software was recompiled for it, you could see how fast it could be, much closer to sandy than Phenom in the vast majority of cases, as it should have been.