Surely you would have the SPEs render to the local XDR memory (which has higher bandwidth than the VRAM). Or better still, perhaps, use tiling and have the SPEs render into their local SRAM where there is huge bandwidth. Once the frame was done it could then presumably be copied to VRAM for output (if that is necessary).
Of course, the SPEs alone will never be as good as the GPU in the system. But they could perhaps do a decent job..decent enough for homebrew games and the like, in the short term. I think there's lots of interesting things that could be done with software rendering on Cell, but in the context of a traditional polygons/texturing pipeline, it'll probably be screamingly fast at the former (transform/lighting), but relatively quite poor at the latter (texturing). I think the real question as to whether a port of OGL/Mesa or some other software rendering solution would be worthwhile is how long we'll have to wait for hardware acceleration. If it's only going to be a short time, it may not be worth it (except for one's own education/curiousity)..although it's arguable that it would still be useful even after that, for projects which wanted to mix Cell and RSX rendering. >From: David Seikel <[EMAIL PROTECTED]> >Reply-To: Discussion List for General Yellow Dog Linux User >Topics<[email protected]> >To: [email protected] >Subject: Re: [ydl-gen] No video hardware acceleration on PS3 >Date: Sat, 2 Dec 2006 10:54:55 +1000 > >On Sat, 02 Dec 2006 08:44:30 +1000 Stephen Braithwaite ><[EMAIL PROTECTED]> wrote: > > > Suppose people did get a project together to port Open GL to compute > > its pixels and vertices on the 6 available SPE's... > > Does anyone know enough about graphics cards to know how fast this > > would result. I think the SPE's would be suitable for this sort of > > work, but how would this stack up against a GPU? Whether or not it > > would be worth starting such a project would depend on what sort of > > results could potentially be obtained. > >The main issue will be bandwidth from the SPE's memory to the video >memory, and I suspect that has to go through main memory. The GPU has >direct access to video memory, so it will still be a lot faster. ><< signature.asc >> >_______________________________________________ >yellowdog-general mailing list >[email protected] >http://lists.terrasoftsolutions.com/mailman/listinfo/yellowdog-general >HINT: to Google archives, try '<keywords> site:terrasoftsolutions.com' _________________________________________________________________ Find a baby-sitter FAST with MSN Search! http://search.msn.ie/ _______________________________________________ yellowdog-general mailing list [email protected] http://lists.terrasoftsolutions.com/mailman/listinfo/yellowdog-general HINT: to Google archives, try '<keywords> site:terrasoftsolutions.com'
