Re: On heavy loads, the 4x4 will perform much better as Kentsfield will get to memory through the FSB straw rather than the dual IMC large pipes.
Many "heavy loads" are not that memory intensive, and this is being marketed as a high-end desktop/low-end workstation, not a server. Two memory controllers paired with DDR2-800 is going to be overkill in many cases. It's hard to find many cases other than synthetic memory tests where DDR2-800 performs much better than DDR-400 in dual-core systems. Intel's FSB definitely has some big limitations, but it will still turn in respectable performance in most reviews/benchmarks.
On HPC loads, the Woodcrest sucks wind and that was with 2 FSB straws.
Good detailed information there. On desktop loads Conroe is faster therefore integrated memory controllers, SOI, HT, etc. must suck wind, right?
And 4x4 gives AMD the option of using coprocessors for the second two cores. Physics ones for games, vector FPUs for HPC apps, encryption engines for security apps, etc. Then 4x4 just blows Kentsfield out of the water for performance per dollar or performance per watt.
Yep, that might interest 1-2% of the quad-core/4x4 market over the next several months. Of course there are no gaming physics processors available for Socket-1207, none that will be available anytime soon that I'm aware of, no games to take advantage of them, and no benchmarks showing their superiority over PCIe alternatives. And of course one of the 4x4 CPUs that were bought as a pair would be sitting around as a paperweight.
And like usual, we will find that Kentsfield uses more power on heavy workloads than 4x4 when you take memory and chipset into account.
Really!? Please link to a 2.67GHz Kentsfield vs. FX-74 4x4 review comparing power consumption. I'm sure others would be interested to see a system with two 125W TDP 3GHz FX-74 processors using less power than a Kentsfield system under load. |
