60 million? Doesn't that seem a bit low?
Maybe.
I'm basing my guess primarily on the historical transistor counts required to derive a "value" part. on 0.15, the transistor budget seems to be between 30-40 million. So 60 million seems about right, for the shrink to 0.13, to give the same physical die size as 30-40 million on 0.15. (I believe the gain for going from 0.15 to 0.13, is about 35% increase in transistor per area?)
I'd think 80 million would do nicely to maintain performance and features and yet still significantly reduce cost.
Possibly, but remember that in the value segment (assuming RV-350 is intened for the $100 price bracket), you can't just "reduce cost". You have to reduce it to a target amount, and that is the first priority. ATI will hack off as many transistors as required to meet it's cost point.
Are you assuming 4 pixel pipelines have been dropped?
I think that's one logical place to significantly reduce the size of the chip. One other one might be to drop support for 128 bit FP pipelines. (Maybe limit to 64 bit.) I believe that's still DX9 compliant, but I'm not sure. They will also likely do away with the extra logic required for 256 bit memory bus. I assume RV-350 will be 128 bit only.
I look to the RV250 versus R200 as an illustration of how much transistor budget this might be able to save instead of dropping pipelines...
So do I.
RV250 did drop pipelines...texture pipelines. I don't think ATI has publically stated the transistor count of RV-250, but I believe it's in the 35-40 million range, compared to R200's 60+ million.
Note that while RV-250 has a cut-down transistor count compared to R-200, it could only maintain similar core clock rates as R-200. Because the RV-350 involves a die shrink, I expect the RV-350 could be clocked a fair bit higher than the R-300.
I might guess that RV-350 to be a 4 pipe R-300 variant, 128 bit bus, 64 bit color only, running at 400 Mhz core / 400 Mhz DDR/DDR-II memory clocks. Sub $150 MSRP.
