At the time of its release, the 4870 was 10+% weaker than the top card, and half its price.
Thats more of a marketing mistake by nVidia, and a surprise by ATI, which made it a success.
The 3870 was a severe alteration of the 2900, which sported a expensive 512 bus, you also still had the rops not changed etc, combined with the DX10 debacle, which was also drasically changed etc etc.
The 3870 was everything the 2900 wasnt, save for a few fixes, which was later corrected on the 4870, again, I point to the DX10 debacle.
Starting out as a midrange card isnt good for Intel in that theyve said itll be "the end of the gpu", as they said, or compete on the high end, as they said.
The window is closing on the mid/low end on discrete solutions, leaving only the upper to high end, thats fact, as Haswell and BD will show.
Implementing ATI and LRB is the reason for this, being done SoC.
These solutions arrive within a few years at most, that leaves a tight time window for a great LRB impact, which would do well to make changes in the gaming market to its strengths as a discrete, and that means games we wont ever see by the time the SoC solutions are already here.
The research chip loks compelling, as its TDP may be 80-90 watts on 32nm, and also be ina small package.
If LRB emulates this in positive ways, then LRB has a chance.
I mention the usage of HKMG simply because what it means in the larger picture, as we would all see disastrous results from Intel without its use on their cpus compared to its competitor, adding it will greatly increase the ability of gfx cards in the future, which again, goes towards a time window and perf as well here.
Intel has said qtr 1, but some things are pointing towards late qtr 2, which plays against all this.
The point of the Tera scale and LRB is easy to see, as theyre set to do the same thing, and shows LRB in na positive light by doing so, as the terascale isnt for sale, LRB will be, and doing those things, so yes, its a look here, this isnt what were selling, but its what it can do, much like Fermi, as nVidia says itll be great at gaming, just dont ask, but it will do these things as well, or, a direct correlation as to each'abilities.
Easy to see, and neither is showing game usage, but are heading their respective gpgpu solutions at each other, and its a no brainer to see this as well.
Theyre taking the same approach, talking to the same peoples, in the same forums/groups, then also being defended by some as saying this isnt about gaming, which again, is what Im getting at here.
Anyways, time is the enemy of LRB here, and its that important. Not for gpgpu, but for gaming. The longer its delayed as a gaming solution, same for Fermi, the worse itll be.
Difference is, Fermi will later get the usage of HKMG, whereas therell be no soup for LRB.
The current games, past games, and future games will be done on DX solutions, with LRB having to break that trend, under a tight window, and IMO needs to have a compelling reason for devs to change their past and present approaches, which a mid range card, whether it sells well in a greta price perf ratio is not a great compelling argument for change.
Add in the drivers, the new driver teams etc, yes, its uphill for LRB, no matter what the price, and its positioning in perf is pertinent for its future influence in gaming, which takes a hit down the road with HKMG coming.
So, you may not see it the way I do, thats fine, but to me, its obvious, Intel wouldnt be dragging out the tera scale if LRB wasnt the vehicle to start this revolution, which is on Intels roadmap, we just dont know when