sarinaide :
If your justification on flop is not beating prior generations or cheaper chips of the current gen then look no further than 2011 with benches claiming a i7 3770K to be better than a 3960X in most
Now that's interesting, because the 3770k wasn't released until late April of 2012 and benchmarks were only leaked a week or two before. Also, I'd say not beating your own chips from previous generations or cheaper chips of the current gen is a pretty goddamn big flop.
The 3770k isn't a flop because it's an incredibly fast chip that mops the floor with every other mainstream chip ever made. Now why don't you post a "BD isn't a flop because..." and try not to make me laugh. Here, I'll get you started.
BD isn't a flop because it cost significantly more than the 2500k at launch while performing significantly worse in the overwhelming majority of applications?
BD isn't a flop because its TDP is almost twice as high as competing chips?
BD isn't a flop because the shared FP scheduler cripples performance on the most common workloads?
BD isn't a flop because it offers better value than chips with legendarily poor value marketed to people with more money than sense?
BD isn't a flop because the chips were so bad compared to similarly priced chips that AMD had to cut the price twice in the first year of sales?
BD isn't a flop because it gets beaten in single threaded benchmarks by AMD's own chips from the previous generation?
BD isn't a flop because it gets beaten in multi threaded benchmarks by chips with half as many cores?
BD isn't a flop because Interlagos was barely able to keep pace with Intel's nearly two year old offerings operating at a significantly lower TDP, and was dominated in 100% of benchmarks by the SB based Xeons?
Sure a few benches and games are better optimized to AMD's archaic K2 architecture but most benches were also done October 2011
And again, rationalization of your untenable belief that Bulldozer isn't as slow as
every benchmark ever done has shown. A rational man looks at a pile of benchmarks showing that A is slower and B and accepts that A is slower than B. You're trying to blame everything but A for its slowness. This is commonly known as
denial.
It's not even a question of optimization anyway, it's a question of a flawed design prioritizing server workloads over desktop workloads, dramatic reduction in IPC over the previous generation, a failure to significantly reduce TDP and inability to produce a chip that can compete on any level with Intel's top offerings. Even the Windows 8 threading optimizations (which are the largest performance gain to be had and by a large margin) only had a minor impact on performance, and only in
some cases.
Optimization might make things
better, but no amount of software changes can fix what's wrong with the hardware. If your workload is floating point heavy then BD is garbage and there's nothing you can do about it. If your workload doesn't parallelize well, then BD is garbage and there's nothing you can do about it. AMD's solution of "throw more cores at it" looks good on paper but fails to deliver in almost every way.
Like I said, Bulldozer is good at exactly one thing. Perfectly parallel integer math. This is an exceptionally rare type of workload, present almost exclusively in rendering and conversion software.
While I didn't at any point say better they are certainly not across the board slower than Phenoms, and certainly not 55% slower than Intel as some have said.
I said they lose 50% of their performance in FP heavy workloads, which is true. The modules are incapable of scheduling floating point work on both of its cores simultaneously. Half of your cores sit idle during floating point work, thus you lose half of the performance. Nobody said they were 55% slower than Intel. Bulldozer is 50% slower than itself when presented workloads that Intel's chips have no problems whatsoever with.
I will say IPC is around 12-15% slower
It's actually about 17% lower IPC than Phenom according to these benchmarks:
http://www.pcper.com/reviews/Processors/AMD-FX-Processor-Review-Can-Bulldozer-Unearth-AMD-Victory/FX-versus-Phenom-Perf-0
A nearly 20% cut in IPC is
massive.
and that is against a company with all the wealth and around 6 years working on their architecture.
So Intel making a good product prevented AMD's engineers from being competent and not losing 20% of their single threaded performance? I'm sorry, I just don't follow.
Bulldozer's IPC being lower has not one goddamn thing to do with Intel. It's a flawed design that AMD took from start to finish. Unless Intel was in there sabotaging designs, then you should probably stop trying to blame AMD's abject failure in IPC on Intel.
I was against AMD releasing BD when they did, what was released was a product well below the initial engineering specs.
As a result of the flawed design of the Bulldozer module, not due to being rushed. This kind of high level design gets finalized years in advanced. AMD made their bed, and now they've got to sleep in it. Another year wouldn't have changed anything. You also have to consider how the market changes while they wait. Sure, they could have made Bulldozer way better by waiting another couple years, but they'd still be delivering 2011 technology in 2013, then. The longer you wait, the
worse the chip looks because your competitors have had more time to produce products that don't suck.
Hell, I doubt AMD could have made a 2500k competitor if they'd delayed anyway. Even if they meets all the goals they have for Piledriver, it will
still be slower than a 2500k on numerous workloads because of the underlying flaws in the architecture. The Bulldozer module itself is the problem. For it to perform well you have to feed it perfectly threaded integer workloads, which as I've explained over and over again, are rare for the average user.
Power and heat, while yes that is not changing a great deal with AMD, they are compensating the loss of IPC with higher clocks
Big deal. They failed at IPC so they were forced to increase clock speed, which increased heat. This is a
bad thing. You're basically saying "Yeah, this part of the chip sucks, but at least they went and sacrificed performance elsewhere to make it suck a tiny bit less!" Intel's chips run cooler and faster each generation, AMD's lose performance and run just as hot as before.
I wasn't one that put expectations on BD, so for that I don't regard it a flop, perhaps underwhelming and badly marketed but that is history
So underwhelming performance and grossly misleading advertising to try to cover it up doesn't constitute a flop?
BD was step one in a new architectural direction for AMD so to expect step one to be exceptional would be over ambitious.
How about expecting it to be adequate? Not worse than the previous generation? A step
forward, not backward? Not priced higher than chips which mop the floor with it in practically every test?