london-boy said:
Right, forgot that. Well it's obvious X360 is cheaper than PS3 wrt bandwidth, it has afterall a bit more than half the total system bandwidth than PS3, of course it's cheaper in that regard.
I would say "different". MS spent 80-90M transistors on eDRAM, so that would be a memory bandwidth expense, especially since it led to a two die GPU.
On Sony's part, outside the expense of the extra XDR royalties/premium, the memory chips themselves are not more expensive. If Sony had gone with 2 GDDR3 pools instead it would have been the same cost for the chips (although there would have been the cost of 2 buses instead of one and an extra memory controller). e.g. MS could have gone the route of segmenting the memory pools into 2x 256MB pools. Of course then you have to account for a memory management model that can work with such. Cell was developed with XDR in mind, and the memory controller in RSX was designed for GDDR3. Each has their own memory controller and dedicated memory type. In many ways it is like a PC: Cell has main memory, RSX video memory. It appears NV augmented the GPU memory controller to read/write through the FlexIO (instead of PCIe) to XDR. Without over simplifying it, Cell and RSX already were designed for two different memory types and they are essentially using their own pools of memory--yet since the FlexIO has such an amazing bandwidth it has been setup to allow RSX access to XDR as well.
Obviously MS saved some money (especially in the future) by not going with a 256bit bus. And having 1 memory controller helps, as well as 1 memory type. But at face value the PS3 does not appear to be seem to be increasing the cost 2x for 2x the bandwidth. Where it might bite them is in the future. Having 1 memory type should allow more consolidation. Having 2x the memory controllers, 2x 128bit buses, and 2 types of memory may make price reduction a little harder.
In this regards you can tell the 360 was designed cohesively and with price reduction in mind. Memory controller on the GPU, shared UMA, eDRAM to offload most bandwidth intensive tasks, etc... (As a side note there is also the XPS which allows the CPU to talk directly to the GPU without taking up the FSB and Memory bandwidth). Maybe the one thing MS/ATI appear to have planned for, but possibly initially a side note, is tiling. MS/ATI must have known they would need to support HD resolutions, but HDTVs have been "in the wings" for well over a decade. Having a 30MB eDRAM module is too expensive, so outright fitting a 720p 4xMSAA framebuffer into the eDRAM was not possible. Enter tiling. But it is pretty clear a 480p image fits in fine, and the troubles (at least initially) if tiling with XPS or Memexport seem to indicate maybe these features were designed for the 480p resolution. Of course maybe marketing just took over with the "720p standard" and/or "MSAA standard at 720p" bit, and of course the delays in working silicon obviously did not help much either.
It would be absolutely fascinating to know how much things changed and what went on behind the scenes during the development of both the PS3 and Xbox 360. When was 720p with AA decided to be standard? Why did NV refuse to create a new GPU design? When was NV chosen for the GPU? Who decided on pricing strategies? What ideas got tanked?
It would be very interesting to learn who presented what ideas and why certain ideas won out, and others did not. My guess there is a lot of back door politics. Sadly I don't think we ever do get the whole story on this stuff. I know we have the Cell story about STI and the Xbox Uncloaked, but those seem to leave out all the really good insider stuff.
