Originally Posted by CravinR1
AMD play all current games just fine, and as BF4 illustrates; are better at future games. The 3570/4670k are great cpus, but in all honestly I quickly see a FX 6300 or 8320 doing better as games are more and more multithreaded to take advantage of the additional cores AMD offers. Continuing to recommend a quad core intel over a 6-8 core AMD reminds me of when people were recommending the E8400 over the Q6600 because the E8400 overclocked higher and no game will ever need more than 2 cores .... right ?
A haswell CORE has approximately the same execution resources as an ENTIRE PileDriver MODULE. The FX-6300 will never be better than the i5-4670K, and the FX-8320 will never be better than the i7-4770K. The Intel chips cost twice as much because in many ways, they are twice as good, which means that all of these chips can be value leaders in their price segments, but there is a very good reason AMD is only getting $120 for an FX-6300 while Intel gets $240 for the i5-4670K...
A single hawell core has 8 execution ports and a decoder/scheduler that can issue up to 4 instructions per cycle. An entire
PileDriver module has 8 execution ports and a decoder/scheduler that can issue up to 4 instructions per cycle. Huh? wait? what? Yea, go back and read that again.
Combine the fact that a Haswell core has access to the same execution resources as an entire PD module, with the fact that it is strapped to a faster memory controller with lower misalignment penalties, higher bandwidth lower latency cache, and better branch prediction characteristics, and those resources can achieve higher useful saturation rates (less wasted cycles)*.
When PD was up against Ivy-Bridge, things were a little different. Ivy Bridge only had 6 execution ports per core. A PD module still had access to more execution resources than an IvyBridge Core, and that could mean a win in parallel workloads where the Ivy's better memory controller/cache/prediction characteristics weren't enough to overcome the raw access to compute resources. When it was Ivy vs PD, PD could be clocked to competitive performance because it still had some architectural advantages. Haswell has dissolved those advantages. The ONLY thing that a PD module has over a haswell core in terms of raw instruction performance at this time, is the ability to clock higher. The problem is that, the ability to clock 10-15% isn't enough to overcome the other advantages listed above (*).
So now that we have established that Haswell CORE and PD CORE mean very different things (with a PD core being almost precisely HALF the width
of a haswell core) we can get past this "MOAR COARS" = better myth. As far as desktop workloads are concerned, having twice as many cores that are half as good is never
a better thing.
When compared 1 thread per CORE (intel) vs 1 thread per MODULE (AMD), Intels approach still has access to the entire 8 execution ports and can still execute up to a theoretical 4 instructions per cycle, the only penalty vs having 2 threads per core is loosing the ability to issue FPU operations on the same cycle as an integer instruction. This is why the i5 and i7 perform basically the same in workloads up to 4 threads. AMDs approach here, cuts the available execution resources in half when the thread count is cut in half, only 4 execution ports can be used when 1 thread is run on a module, this limits compute throughput to a theoretical maximum 2.6 instructions per cycle.
My cell phone has more cores than an i3. Doesn't make it better.