Yet you keep bringing up JF's comments which in every thread I've ever seen he always says "these comments are about server performance".
Also you seem to not know what IPC is, separate the words, instructions per clock, NOT performance per clock, NOR performance per CORE.
Even then that's open to interpretation, because it can decode less instructions per clock but the core itself, despite one less integer pipeline, can actually run through more per clock. In terms of server throughput, theoretical instructions per clock and ACTUAL instructions per clock are VERY different, you can reduce the former AND raise the latter, realistically you'd end up with a better performing chip if you lowered peak but raised the average performance.
But again, you have said JF/lies/AMD/marketing multiple times on this page alone, and you use JF's statements which are specific to server chips, then ask everyone to ignore server performance because only desktop matters.
Then we can bring up things like, cache being broken(if it is, or poorly implemented) can drop effective IPC while the core is no different. Its highly likely the SAME cores with the SAME clock speed with different cache could be significantly faster, it seems obvious to everyone but you that IPC can be dramatically improved with proper scheduling. Nothing so far has disproved that IPC on Bulldozer is higher than Phenom, but a factor of MANY things outside of the core itself can lead to lower performance, there are situations in which it is faster. If they bought out a fixed bulldozer with fixed cache but identical actual modules, and performance was up 30% the effective IPC of the modules/cores will not have changed, just the overall package will perform differently.
You change your argument to suit, I'll use JF's server performance claims about desktop and bemoan anyone else mentioning server performance as its not relevant.
You can't prove IPC is lower, you can't find JF say he talks about anything other than server performance, server performance is relevant just not to you.
This is not an accurate statement, at all, what you I assume meant to say was, performance per core improvements inherantly increases performance across the board........... IF you make no sacrifices in the architecture to raise per core performance.
In reality AMD could not fit 8 "wider" cores on a Bulldozer die, its that simple. Unlike you I have from day one said, per core performance IS irrelevant. There isn't a single thing I'd do on a 2500k, 2600k or Bulldozer, or Phenom x4/x6 where single threaded performance matters any more. There isn't a single game I own, going back 15 years that is single threaded that won't run great on a Bulldozer, new games are all multithreaded, some of them are heavily multithreaded, performance in all of them is more than ample, and going into the future more games will be more heavily multithreaded thereby using the bulldozer more effectively.
Anyway I went a little off track there, to raise IPC as you see it, performance per core on a single thread, you WOULD need a more robust front end AND another integer pipe, maybe a little more FPU aswell, this would all have made it impossible to have 8 cores on a die. So you'd end up with 4 "fat" cores, faster in single thread, same speed or even maybe slower in heavily multithreaded applications. Heavy multithreading IS the future, bringing a new architecture out now, which will carry over for the next 4-5 years at least, and focusing on single threaded performance WOULD end AMD, its that simple. Basing an architecture that needs to be competitive in 5 years, on single threaded performance, would be beyond stupid.
In your magical world of make believe where you can make the cores wider with no die size penalty and increase the single threaded performance, sure, that would be great, but we don't live in the land of Martini makes up the rules of physics.
I really have no idea what you are getting at with this, it seems to be not the first time where you suggest a 10-25% performance bump in single threaded performance is laughable as it won't improve 8 threaded performance.........
Yet in any other post all you complain about is single threaded performance, and randomly ignore the fact that with 8 threads Bulldozer beats the 2600k multiple times and the 2500k almost every time. So the one area it would bump performance is the one area you complain about but this is somehow.... laughable.
More importantly, with a 15-25% bump in single threaded performance in 4 or less threaded performance, Bulldozer would likely beat the x6/x4 in almost every single situation with mostly similar clock speeds.
Lets be honest, you talked crap about Bulldozer when you didn't understand it, you talked crap and literally lied about what other people had supposedly said about it ( You have indeed accused me of saying Bulldozer would be awesome and be faster than Sandybridge, when I was infact one of the few to point out exactly why it would not), you're still talking crap about Bulldozer even ignoring all facts now known. You consistantly call it slower than an X6/x4/previous gen, despite multiple situations in which it trashes the x6/x4, which are, as it happens, area's where software will be going more frequently in the coming years and, meh, not much else needs to be said.
You've been trolling this thread since the start, posting more ill informed crap than anyone else, doing a bit of a "raven" by trying to hype expectation about what you think it should do, then coming in after and saying how crap it is and you seem to have a fundamental lack of knowledge about errm, everything in which you talk about.
they did indeed claim early on that their intention was to hold the line with IPC with regards to Phenom II, but JF-AMD has been saying toward the later part of the development process that IPC haven't only stayed the same, but improved over the Phenom II series.
