Tyree Walker,
Core count and clock speeds are only useful comparisons within the same CPU architecture.
Haswell and PileDriver are not even remotely similar in terms of execution resources per core, in fact, a haswell core, has more execution resources, on a shorter instruction pipeline, with better cache bandwidth and latency, than a piledriver MODULE. Yes, 1 haswell core is more powerful than 2 piledriver cores. More transistors, arranged better.
Hyper-threading is not a technology that makes a core more powerful, it's a technology that lets us get more use out of an already powerful core. A hyperthreaded haswell core and a PileDriver module, both have dual threaded front-ends. The key difference, is that in the hyperthreaded core, any single thread can gain access to ALL of the execution resources in the core. On the PileDriver module, any single thread is confined to be executed on only some of the execution resources in the module (1 of the 2 small cores). In real-world workloads that can't always be made to scale evenly across many threads, the haswell core proves more powerful than a piledriver module, over and over again....
When compared clock for clock, 1 thread per haswell core (hyperthreaded or not, doesn't matter) vs 1 thread per piledirver module, haswell is ~75% faster.
When compared clock for clock, 2 threads per non-hyperhtreaded haswell core vs 2 threads per piledriver module, they perform the same.
When compared clock for clock, 2 threads per hyperthreaded haswell core vs 2 threads per piledriver module, haswell is ~25% faster.
What this means, is that when we compare an i3-4150 to the FX-6300, the i3 winds up performing better in any workload that presents as 1-4 threads. It isn't until we find a workload that achieves high continuous saturation on 5 threads, that the FX-6300 finally catches up to the performance of the i3, and only when we find a workload that achieves continuous saturation on 6 threads is the FX-6300 finally pulling ahead of the i3-4150 by ~25%. The turbo speeds of the FX-6300 only help mitigate the lightly threaded performance advantage of haswell down to ~50% (from the clock-for-clock advantage of ~75% mentioned before).
In real-time workloads like gaming, it's not actually possible for the compute workload to present as a perfectly even amount of work across many threads, there are too many variables in workload for each job and all are constricted by a timeline. Typically only 1 or 2 threads at a time can ever achieve really high saturation simultaneously in these workloads. The result, is that even though the FX-6300 can technically surpass an i3-4150 in raw execution throughput, the i3-4150 consistently out-performs the FX-6300 in games.
--------------
-------------
Jason 0115,
There are lots of considerations beyond just the raw execution performance I have described above that should be taken into consideration when selecting a CPU. I think it would be prudent to consider what you intend to use your CPU for, to determine what budget CPU is best for you.
For example: The Pentium has support for several key instruction sets DISABLED, which will hurt it's performance significantly in some software.
The AM3+ platform, offers key advantages like ECC memory support, and an IOMMU that you won't find on anything from Intel until ~$200 for a CPU, and ~$200 for a motherboard.
The 760K, when compared clock for clock, is comparable to the G3258 in execution performance, but spread out over 4 threads rather than consolidated in 2 threads. Choosing the best among these 2 is going to be workload dependent.
Facebook
Twitter
Instagram