leechan25 said:
'Based on' is a loose term. Could mean G70 with a few bells and whistles, and could mean a new architecture with nVidia style shader ALU's combined into SPEs. From the info we have and the fact the nVidia toolchain is in use, which suggests very strongly they're not rewriting it for a totally new architecture, I think RSX follows the nVidia archtecture of separate vertex shaders and pixel shaders, and there will be no SPE's on RSX.leechan25 said:
There's no official word beyond what we've already heard that points to something traditional and conservative. But by the same token there's nothing to support strongly a view of something radically different, plus a history of patents filed that never amount to anything to show a patent isn't reason to expect anything. If you want to discuss the patent, it's been discussed before and so isn't anything new. If you want to propose this is a patent and patents are strong indicators of likely products, I say otherwise and that a patent is nothing more substantial than the recording of an idea someone had. I'm not really sure where you want this discussion to go. If you want ideas on how good or bad this architecture might be, do a search. I'm sure that discussion's here somewhere. Checking back I see XBD's linked to the probable discussion. But I doubt you'll find many people who'll believe this represents what RSX is. We may all end up being very surprised, but the logic behind believing in something exotic isn't there. If you want to make a case for this technology, do it by discussing the merits of using SPE's over Vertex Shaders, cost effectiveness, portability of nVidia's existing toolchain etc. For me, not being a low-level GPU programmer, I don't know where you'd want SPE's instead of Vertex Shaders, especially if you can't clock them high. Seems to me a lot of research to come up with a part that'll do the job of nVidia's existing part and cost more to do it. The whole idea of the Cell GPU was a unified architecture for scalability IMO, a one architecture fits all solution that can handle anything thrown at it. Performance and cost effectiveness and developer friendliness probably isn't there ready in time for PS3. So I don't know that this architecture would be better than a G70 varient, which would explain why it may have been dropped (if it was ever considered, and if it indeed has been dropped!)
Sethamin said:
cho said:
wireframe said:I don't think this is a game of Poker and especially not about a bluffing game unless you want to see it that way. It's about execution, cost management and talent/contract management. I'd liken it more to a game of Risk or Monopoly
wireframe said:
I'm trying to remember if I thought RSX had a 256-bit bus, b/c a 128-bit bus feeding 32 G70 pipes sounds a bit wee. Doesn't it? I'm thinking 32 potential TMUs competing for 128 actual bits.cho said:
wireframe said:What makes this most ironic and bizarre is that Nvidia supplied a very "generic" NV25 derivative for the original Xbox and that console is often praised as having the best graphics of its generation. Why does everyone seem so worried that RSX "is nothing special and magical"? Sure, things may have changed. But more so than ever, graphics are becoming a specialist field and there are two main players who know how to do real time 3D: Nvidia and ATI. So what are you so worried about? I get the feeling people would worry less if it were NV30 based as long as it had some eDRAM.
