I think that Barts contain 1280 SPs.
Yes I know very well that this is a "high level" overview but it seems to work.
Althou i'm not so good at math you're right there, and it aint
just a high level overview. But then EG_Redwood and EG_Juniper have only one setup engine to oversight their work, while EG_Cypress and NI_Barts have two, so it's more proper way to me looking what is cut down.
They're just cut down EG_Cypress part, no "new NI shaders" (that took more die space per SP), all tessellation improvements in Barts came plainly fro it's higher clock so there's no real improvement (or need for die area) and with all stuff they cut-off (320SPs, DP support, 2nd CFX and reduced MC) ... it simply should be 1280SPs
Question is: does using a power-of-two SIMD count use that much more die space than a power-of-two-minus-one count?
(...)
Is it better to have coarse grained redundancy with a merely bigger die, fine grained redundancy or no redundancy at all?
I agree there with you. Redundancy if it translates into merely 15-20mm2 (7-9% bigger) die space is nifty feature for gpu business.
And who advocated that 1280SP "new shaders" require 280mm2?
Lets hope that nVidia will soon introduce speculating gtx475 cards on fully working gf104 chips, and that amd will be forced to abandon 6850 with laser-cut 160SPs in favor of fully working HD6870.
What makes this chips so large?! It doesn't even support FP64, nor even have 1280SPs, but accordingly it sports somewhat smaller Redwood MC (x2 to become 256-bit, and not supporting differential gddr5 signaling makes him 50% smaller than one in Cypress/Juniper).
So what's the legend behind it? It's not like only one more tessellation dispatcher aka. RPE (what it stands for?) needs to consume enormous die space.
but seems nobody doesnt wish to confirm my thoughts
Large considering only 70% instead 80% active engines (ed:compared to Cypress)
As now seems, neither AMD doesn't want to address fully working gf104 chip, that is still non-existing as real product ex.gtx475->now.gtx560(?)
And as the only reason for redundancy in Barts would be to address some mythical future nVidia product, so as there are no requirements for it we probably wont see that
"fully enabled Barts" either. But who knows may we'll see
HD6890 after all and that would be great
I'd bet on a manufacturing or driver issue.
My bet is on AMD cleansing out of piled up HD5800 cards from clogged warehouse inventory and further down the retail stream before newcomers arrive. It's not like they're really afraid vaporware GTX580 will spoil their sales
And Q3 show some sales slowdown-drop of ATi gpu line so its last moment to sell them at discount prices and yet for none the less than HD6800 cards.