Pip we had this discussion and i cannot disprove what you are saying,but do you honestly believ the l3 on am2+ will be the performance raising feature? AM2+ may have a dual Imc as well,which is not an am2 supported feature.
lets go point by point over the differences.sse extensions will give what level of perf?And the real estate for them will still be below intel.
How will L3 ,and dual imc and the added extensions accomplish what you are implying on am2?
Let's not confuse AM2+
socket with AM2+
processors.
AM2+ as a socket offers very little for performance, if at all.
Now, concerning the improvements of the AM2+ processors (K8L core):
L3 cache? A little, but mostly for 4 core systems. It helps with core-core on die communications
Dual IMC? I'm not aware of that.. AFAIK, a single CPU will still have only one IMC.
Enhanced crossbar switch? Again, will help with core - core communication, but it's not a major point here
SSE4? This will help for future applications, but it's just to match the instruction set of C2D
Improved branch prediction? Will help especially with certain types of integer code. Hard to quantify, without more details about branch prediction hit rates on K8, C2D
Out of Order Loads? Again, will help with certain types of integer code. (non computationally intensive)
Improved TLB? Will help with server / scientific workloads (large datasets)
So where is it the big performance improvement of K8L?
Simple, they basically
doubled the width / throughput of the main computational engine, SSEx.
This involves doubling the execution resources (ALUs), the L1 cache ports (2x128bit Loads, or 1x128bit Load + 1x128 bit Store) and buses, the schedulers and even the fetch size (to be able to decode 3 SSE instructions per clock under all conditions, even in 64bit mode, something that it seems C2D can not always do (but there's not reliable data around)) this is probably the biggest factor of increased performance in K8L.
And crunching numbers at twice the speed (theoretically) they'll finally be able to make some use of all that bandwidth which is mostly wasted on AM2 processors.