The compositing chip is already not used by ATI for SuperAA.
No CrossFire for R4xx?
- Thread starter Geo
- Start date
Re the separate compositing chip, and integrating the logic into every R520 --we do have ATI's testy notes response to the XFire pdf (from "where ever"). It seemed to deride the idea of making everyone pay by putting the logic inside every gpu. But then that wasn't meant to be public, and was clearly "blowing off steam" rather than thotful forward-looking.
Personally, as I've said many times before, it seems to me that the NV implementation of not requiring a special master card makes a lot more sense for partners, and I think ATI (and, us) would be well served to mosey in that direction.
Personally, as I've said many times before, it seems to me that the NV implementation of not requiring a special master card makes a lot more sense for partners, and I think ATI (and, us) would be well served to mosey in that direction.
At the moment SuperAA is the only mode that doesn't use the compsoite engine - the data is passed from the slave board to the master via PCI Express and blended on the graphics chip. It is intended that at sometime in the future SuperAA could/will use the composite engine for a bit of a performances increase, but currently it doesn't.
trinibwoy said:
Reasonable point. In theory, you could argue those transistors are paid for one way or another. . .but in practice this has meant a cost disadvantage to NV (bigger chip) rather than a price difference to the consumer.
I'm really beginning to wonder if NV has something fundamental figured out on yields that has ATI puzzled since NV40. Looking at die sizes and transistor count "mysteries", greater availability of NV parts initially, the low power draw of G70. . and that NV is doing verywellthankyouverymuch financially with bigger dies at the same price point. It isn't supposed to work that way.
geo said:
I dunno. I think it has worked like that for a long time and just recently ATI and Nvidia thought they would get real clever about how they went about the actual manufacturing and it seems to have bitten them in the behind. I think most of the IC market is discovering that "don't get cute" is a good mantra because at the end of the day you need a product. It's one thing to have a testing lab/group design something that will revolutionize the market (throw several millions at layout and process techniques and see what comes out) and quite another to base your bread and butter on it.
I think it is this promise of getting it done even cheaper that has bitten into many companies and they begin staring themselves blind at improved margins (earnings) when they really should be happy with the tremendous markup they already enjoy and focus on getting solid products out the door.
Clocks may have something to do with it too. Jen pointed at that at one point as a philosophical difference between the companies recently, and he felt theirs was much better for yields. Arguably that helps to cancel out your cost disadvantage on die size.
Looking even just internally at ATI, the diametrically opposed yield results of C1 and R520 right now offer another data point in support of that rationale.
Looking even just internally at ATI, the diametrically opposed yield results of C1 and R520 right now offer another data point in support of that rationale.
