cummy :
I like the AMD lineup and I am wondering if anybody has thought of what a Phenom 3 might look like? would it be worth it? Yeah and I know it is not on the roadmap.
I would like to see AMD experiment with a Phenom 3 processor. First of all a die shrink to maybe 32, 28 or 22nm would be a nice start certainly improving Energy output, changing the L2 cache to maybe 6mb and dropping the L3 to about 4mb; leasing the hyperthreading technology from intel and that would be something interesting; raising HT if it is possible. AM4 anyone?
I'd love to hear if Phenom could get a third life or this is really it.
K10/Stars/Phenom sort of did get a third life as the "Husky" core in the Llano APU. It's really more of an Athlon III as it lacks L3 cache, but it does have 4 MB of L2 cache for a quad-core chip and is an improvement over the old Propus Athlon II X3/X4 die. Llanos do fine at idle for power consumption; it's the load power consumption that's not so stellar, mostly because the default Vcore on just about every standard-TDP AMD chip since the 65 nm units has been absurdly high.
Raising HT speed on a single-socket CPU does little to nothing for performance as all the HT link does is connect the NB (and indirectly, the SB) to the CPU. Not all that much data flows through there unless you have a non-Sideport IGP in the northbridge. Some people post that raising the HT speed helps performance because the HT speed depends on the uncore (L3/IMC) speed and raising the uncore's speed really does improve performance (mostly due to higher L3 speed) especially when there is a large difference between core and uncore speed.
If you wonder what HyperThreading would look like on an AMD chip, look at Bulldozer and you'd get a general idea. Bulldozer's multithreading is a little different than Intel's HyperThreading as each logical core does have some of its own dedicated physical hardware in Bulldozer (integer pipelines, an L1 cache), but it follows the same general idea of scheduling two threads per one complete set of core execution resources. Intel's approach is quick and dirty but cheap, AMD's is a little more costly but more elegant and has the possibility to perform much more consistently...once the likely cache-related bugs get worked out.