Originally Posted by Cyro999
No it doesn't, i don't have a bunch of numbers but fx@5ghz scores ~1.35 in cinebench 11.5 while you can exceed 2.1 with Haswell. Some say cinebench 11.5 is somewhat biased, but even with the performance penalty from running both cores in a module, piledriver is nowhere near haswell in terms of IPC, the gap is like 50-60%. It's expected to be worse, considering the module design and the fact that unlocked 8 core piledriver is significantly cheaper than intel's lowest end unlocked quad core. 30% higher IPC and other stuff steamroller brings would be godly though, 20% would be awesome.
Ivy Bridge is, at max, 55% faster in IPC than PD going from memory.
Originally Posted by PureBlackFire
If AMD can even get +/- 5% the ipc of sandybridge and have low tdp quad/hex core cpus for $300 or less that would be a good enough progress with this architecture.
The IPC an AMD CPU offers now is enough for >95% of games out there if you're using a single GPU.
Originally Posted by AlphaC
Floating point is even worse than bulldozer, so unless the floating point can be shoveled off the the GCN cores (GPU basically) for processing I don't see the hype
First off, the source mentions the lack of L3 cache as a possible reason for that.
Secondly, FP is by far in the minority of code ran on a modern CPU.
Originally Posted by pokerapar88
Been there, done that, with an i5
No, you haven't. My i5 is (slightly) faster than yours due to equal clocks and slightly higher IPC yet a FX-8350 can easily take it, or a 3770k depending on the benchmark.
Originally Posted by Cyro999
It's common knowledge that Haswell is like ~50-60% faster core for core when you close to match frequencies though (average oc on piledriver high end is only 200-400mhz higher than haswell)
Piledriver runs a module for every core that Haswell has to stay at all competitive. An 8350 with half of it's cores disabled would get laughed at by an i5 - it wouldn't be a +30% lead for i5. They have four modules and 8 threads for a reason, they need them and it's completely expected for them to lose by a margin greater than 30% when half of their cores are disabled.
400Mhz is more typical, remember that the FX Piledrivers don't have RCM or whatever is in the A10-6800k that allows it to hit 5Ghz left and right while a Haswell can vary from matching it to barely going above 4Ghz.
That entirely varies on the program you're talking about...There's benchmarks where an FX-8350 matches an i7 3770k and people seem to forget that IPC varies from app to app. (Hence why Ivy Bridge can vary from 0% to 55% faster in IPC than PD, 0% being only in a handful of apps with most being around 30% and a few, very poorly threaded programs being closer to the 55% mark)
If this 30% is normal, expect SR to be slightly behind Haswell with an equal amount of threads in most apps, well behind in some and ahead in others if they launch an 8 core AM3+ one, either way this might finally end the completely unnecessary AMD hate...People don't seem to realize that Piledriver, Deneb, Nehalem and even the better clocking Yorkfield CPUs are fine for games if you're running a single GPU, or an older one. Sandy, Ivy and Haswell really only are useful when you're playing highly CPU intensive games like Starcraft 2, Sins of a Solar Empire, Civ V, etc or are planning on multiple recent GPUs.
(Even then, my HD4890 CFX ran perfectly fine on my Phenom II x4 @ 3.6Ghz and wasn't far behind results I saw from people with i7 920s and the like, but I can't speak for modern R* 2*0s and Keplers with that.)