Provantage shows the 95w 1055T as special order and I have not seen any online vendor with the 95w version in stock.
Availability does seem a bit poor on these chips. It'll ship, it'll just take 15-30 days to get to the buyer, unless ProVantage flubs the whole thing up. Kinda sucks but oh well.
There are no process or stepping differences, it is all just binning for lower Vcc at stock clockspeed as you stated.
Figures. But yeah, that makes sense.
The thing is, and this has been verified to me offline by both AMD and Intel engineers, that if you want higher clockspeeds then you actually want the chips that have higher leakage at idle (higher static leakage, Ioff and IDDQ) as these are the chips whose transistors are all setup for quicker transition from the off state to the on state and that means higher clockspeed potential.
I have heard that as well. At the same time, that fact does not (necessarily) seem to affect chips that have a lower vid than others within the same microarchitecture/process/stepping/etc.
In some cases, the chips with the lowest vid are the cream of the crop. AMD's last foray into binning for low vid/vcore was the Athlon XP-M 2500+, if I recall, and we all know how that worked out.
Their 'e'-designated CPUs from AMD seem to be an entirely different animal that have transistors that are deliberately set up for lower leakage; in other words, I don't think those are the product of mere binning, though I could be wrong. Regardless, the 'e' CPUs are pretty poor overclockers if I recall correctly.
Cool. How much control do you have over your ability to select which two cores are specifically disabled or enabled? Does your logical core selection always correlate to a physical or does it float around? Can you find out the weakest core by process of elimination and likewise your best clocking cores and further optimize the overclock?
I think this varies based on the board. Back when I had my x4 635 working, my 790FX-GD70 would allow me to disable any core, and allegedly would let me increase or decrease clockspeed of any core by increments of 2% over the current clockspeed (though it never seemed to work right; that feature might have relied on the chip being a BE or something). You can get finer control of individual core speeds using something like K10Stat I believe.
Figuring out which cores would clock higher than others is a bit trickier, though usually it's one core in particular that likes to fail when there are memory/NB stability problems. I never OCed on a per-core basis (never got around to it before killing my chip), but it could be done, at least with K10Stat. Detecting which core(s) were most prone to failure during moments of memory instability might be a good path to follow as well.
I've gotten some DDR3 sticks to play with so I mated them with an 1090T, and learned the performance reliance on memory configuration can be shockingly high at least under some workloads.
All I can say is: awesome. Thuban's IMC is pretty impressive from what I've heard, with DDR3-2000 being quite possible with the right memory.
I am not surprised to see notable performance differences at higher NB speeds, though the amount of difference is quite striking. I suspect that NB speed will make a bigger difference on K10.5 chips with larger core counts. It would be fun to re-run all those NB performance tests I did on the 635 using a Thuban chip. Well, maybe not fun, but informative.