Ignoring the part numbers the line transitioned as had been usual for Intel with increased clockspeeds and then the addition of more cores. 2.66x4, 2.93x4, 3.2x4 at launch.
Core count doesn't not equal architecture. The 4 core count was left behind in architecture. There were only Nehalem 4 core count model. There were zero Westmere updates. And no Westmere was not solely a process shrink. It wasn't a major micro architecture change there was some ( AES streaming and some virtualization updates ).
The 3500 series on the at this point relatively ancient 45nm really couldn't be cranked much higher than the 6 core Westmere models which is what should have been done.
That sequence of clock speeds exactly
backwards of what the sensible progression should be. The 4 cores should be the fastest and the 6 core trailing behind or equal. Indeed the sequencing now is:
E5 1620 3.6 GHz ( 4 cores )
E5 1650 3.2 GHz ( 6 cores )
E5 1660 3.3 GHz ( 6 cores )
For those looking for single threaded drag racing solutions the 1620 would gap all of the iMac and mini solutions. Even likely the upcoming Haswell upgrades for those two in the Fall.
I never saw this as quirky, I'm sort of surprised you do.
Not sure why.....
So they didn't use Westmere to make, for example, a native 3.6GHz 4-core CPU like they have with the E5-1600 line. But then didn't make any native 4-core Westmere CPUs, and the 4-core models they did make are 6-core with 2 disabled and not impressive over their 6-core brethren.
Because a 45nm die with 2 cores switched off is cheaper to produce in volume than a 32nm die with 2 cores switched if the wafers are the same size? Not really.
I would venture to guess they also saw a major shift in that LGA 1156 was much more popular than LGA 1366 after its launch,
Maybe, just maybe if the 32nm facility to filled to capacity doing mainstream design perhaps. If there is a too few items to crank out on 45 and 32nm has capacity problems but otherwise now.
More likely it was a designer resource constraint and some folks pulled off jumpstarting a real 3600 line to get the E3 line cranked up.
The sockets aren't more popular as much as the far lower TDP points that the E3 line brings to market. Up until then Intel had tried to fill the "lower power" server CPU market by just completely gutting the clock speed. That isn't as effective as something that is just designed to run sub-90W in the first place.
I don't recall any outrage that if you paid more you got more cores, I do however know people were critical of the current line up where you get less clockspeed if you want more cores and you get less clockspeed and performance per clock with 6-core CPUs that the consumer line.
From the clueless "just more GHz" crowd? Probably so.
For round numbers lets say the memory bandwidth coming out of the package is 200 GB/s.
If each core needs 30 GB/s at 3GHz and 40 at 4GHz then a 4 core model ( 4 x 40 => 160 ) will fit better at higher speeds than a 6 core model ( 6 x 40 => 240 ).
I can see way folks not paying any attention at architecture would complain. The higher GHz has more crotch grabbing bragging rights. It doesn't necessarily lead to increased performance.
The E3 line replaced the Xeon 3400 series, which was the Xeon version of Clarkdale/Lynnfield and replaced the Xeon 3000 series of LGA 775 processors.
If skip back past the 3600 and 3500 where it was merged up. During 2010 there was a gap. E3 didn't appear to 2011. But yes elements of the 3400 line up came after the 3500 series shipped. There was two usages of the 3000 indicators to outline really different line ups. Sort of how the Core i7 is substantive muddled now. In a Mac Pro context though those "desktop" variants were never used.
Why should there be a big gap between an entry Mac Pro and iMac when both have a $330 CPU in them?
I didn't say big gap; just gapped.
If the Mac Pro's processor is not burdened by an iGPU there should be some gap. If there are enough extra transistors so that have substantially higher I/O bandwidth ( 40 lanes versus 16 in current set up and about twice the memory throughput. )
