Quote:
Originally Posted by
S.M. 
Because K10 and K10.5 isn't better than Trinity. So likewise worse than Vishera.
That's probably why. Faster, thermally smarter, efficiency scalable...etc.
You honestly think AMD looked at K10 and went "Huh, oh man, look at all the vast potential left in this architecture." And built something else? A simple die shrink wouldn't keep AMD relevant and it wouldn't win AMD back it's market share. Might as well engineer something while you have money.
The Bulldozer archtype is to combat hyper-threading. It's not a case of "moar coars" but smaller physical cores for cluster multi-threading instead of logical.
I'm still literally
astounded that people on this website think AMD doesn't know how to run a business after front-running it for years. You honestly think you know more about K10's architectural potential than AMD?
this is really about software in reality. Most software has a useable IPC of around 1.7-1.9. Which means the phenom 2 ipc front end width at 3 was just being dramatically under utilized. This is what hyper threading is really about btw. So Intel makes the 4 wide scheduler with a wide pipeline, and then decides hey, we can just shove more stuff down this pipe since its so wide.
except when you have contention for memory and resources.
So what AMD did was realize that most workloads are integere, and increase integer resources, they just left thefront end a bit to narrow the first go round. They also get better thermal loading across the chip, and they also know that at some point, clockspeed must go up since software seems to be pretty much stuck with a ipc of 2 give or take.
So bottom line, AMD optimized the design, they just didn't catch the bottle necks up front. And synthetic benchmarks run with a IPC of greater then 2, which favors intels design, but most apps don't make use of that much IPC.
Its actually a pretty smart bunch of compromises, what they should have done is made the whole chip adaptble to handle wide ipc needs when requested. That would have really helped the benchmarks.
Just thought you should know. 
