Originally Posted by KyadCK
Find me even one program that can use 6 cores but not 8, without you putting limitations on the program, and your argument will be legitimate. If only for that one program.
Irrelevant, as most that can use 8, can use 12, and so on and so forth, until we're comparing clusters... As soon as the bar has been moved to accommodate the additional parallelism, then ALL considerations of increased parallelism are on the table, otherwise it's a double standard. Being ABLE to use more cores to take advantage of the compute efficiency afforded by parallelism is wonderful, but it doesn't answer the question of which core design has better compute efficiency.
Until then, adding more cores at slower speeds is a way to increase computing efficiency
Exactly my point all along. If parallelism has the potential to improve compute efficiency, in and of itself, then taking advantage of parallelism will mask architecture differences and make it impossible to distinguish which design has better baseline compute efficiency.
since a stock 8350 will perform the same as a 6350 at 5.3 in multithread while consuming far less power. That is how the real world works.
Yea, a point that I have repeated dozens of times now. Finding out that parallelism improves compute efficiency is not the point, everyone knows that.
Furthermore, an FX-6350 and FX-8350 are from the same architecture, so they have the same compute efficiency when tuned to the same parallelism and clocks. This thread concerns Phenom II vs Vishera. If I want to know which architecture has better compute efficiency, and I mask the difference with a difference in parallelism then I won't have an answer to the question.
The fact Vishera can get up to 8 cores on a single die is a part of the architecture,
The core-count limitation is the socket/platform, not the architecture. Parallelism can be approached in many different ways. More sockets, bigger sockets, etc. G34 and 2011 are examples of platforms that use the SAME CORE ARCHITECTURE as the "consumer" counterparts with a lot more cores on the same die. The architecture does not define the maximum parallelism possible for a single die.
so crippling one arch by forcing it to limit itself to 6 cores doesn't tell you anything.
Parallelism is not "part of the architecture." It is a consequence (or advantage) within die/socket constraints.
You can still compare perf/watt, single thread perf, and multi-thread perf regardless of cores.
Yes, as I have said, we can compare as many cores, vs as few cores as we want, but the results of any such test will not answer the question I have.Edited by mdocod - 3/5/14 at 11:57am