Can the RSX framebuffer be split between GDDR3 and XDR RAM?

[Fafalada]

As I said before, if the texture cache is working properly then bandwidth might not even be a problem for post-processing, especially with these long shaders that nAo and Fran are talking about. I can understand why XB360 didn't go the PS2 route.

Except addressable eDram can be useful for a heck of a lot more then just post-processing.
PS2 was just the tip of the iceberg (since it was limited to storing only textures in it) - which is why I mentioned PSP as 'ideal' config.

And Xenos flexibility could take that to a whole new level (with memxport it would become a localstore of sorts) - you could dynamically generate any data from textures, geometry to display lists, inside the GPU, without even touching external mem. Not to mention being able to run non-graphic processes in GPU with almost no external memory references.

Indeed so. We programmers are historically very bad at guessing bottlenecks in what we write without actually profiling it

Indeed, there's that saying "any optimization you write is not an optimization until you profiled it to show it really works".

 

[DUALDISASTER]

Except addressable eDram can be useful for a heck of a lot more then just post-processing.
PS2 was just the tip of the iceberg (since it was limited to storing only textures in it) - which is why I mentioned PSP as 'ideal' config.

And Xenos flexibility could take that to a whole new level (with memxport it would become a localstore of sorts) - you could dynamically generate any data from textures, geometry to display lists, inside the GPU, without even touching external mem. Not to mention being able to run non-graphic processes in GPU with almost no external memory references.


Indeed, there's that saying "any optimization you write is not an optimization until you profiled it to show it really works".

 

[Mintmaster]

hm... you make it sound like the addressable eDRAM is something we won't see again. Or... how about having two different eDRAM pools? But I guess that defeats the purpose (post-processing)...

I didn't mean to make it sound that way. Manufacturing was a big problem here. Not everyone can make eDRAM well, and NEC probably can't make a chip of Xenos' complexity as well as other foundries. Addressable eDRAM makes a lot more sense than non-addressable eDRAM in most cases, but just not given these specific design parameters. They couldn't have it all on one chip, and inter-chip communication is comparitively limited. I think I read a slide that said this was one of the most challenging parts of getting Xenos working.

Except addressable eDram can be useful for a heck of a lot more then just post-processing.
PS2 was just the tip of the iceberg (since it was limited to storing only textures in it) - which is why I mentioned PSP as 'ideal' config.

And Xenos flexibility could take that to a whole new level (with memxport it would become a localstore of sorts) - you could dynamically generate any data from textures, geometry to display lists, inside the GPU, without even touching external mem. Not to mention being able to run non-graphic processes in GPU with almost no external memory references.

Oh don't get me wrong, I think addressable eDRAM/ZRAM would be awesome, especially for future GPUs with say 10x the power. I was just talking about it in light of the design constraints of Xenos. The post-processing talk was just the main topic of discussion, that's all.