marconelly! said:
Seriously, though, what is your beef with Gaming Age? I don't know if you think this board is more 1337 or something, but misinformation, trolling and stupidness here, sometimes easily exceeds anything I've seen on GA lately.
There has been an influx of people within the last year that has really.. um..
modified the dynamic of this board. To be quite frank, yes, this board historically has been "elite" among open 'net communities in that it's population is highly educated in a multitude of topics and have the ability to dynamically address and discuss a problem in intricate and often interdisciplinary ways. B3D has been the birthplace of many discussions and ideas which would eventually influence the PC IHVs, granted with the demise of 3dfx the level of "professional" interaction here is not as noticeable, it still harbors many industry members.
For you to compare Beyond3D to GA (Which I occasionally read) is in-itself a fallacy and not exactly how all of us have seen B3D. In fact, I question if the influx of.. "modifiers" isn't due to the awareness of B3D being raised in general, but more specifically the references I've seen at GA.
Well, for what it's worth I don't think they are radically different in their ideology. Cell sounds like a natural expansion of EE parallel computing concept, and I suspect the majority of vertex processing will again be taking place there (on the Cell), instead of the graphics chip
Perhaps similar in that they are both machines which use concurrency to side-step production limitations, but I don't see much more after that.
Where the
Vertex Processing takes place is a non-issue in what Ben's talking about. In fact, when designing a closed console like this, with parts tailored specifically for this function, where the processing is done isn't a problem. "Vertex Processing" was moved by the IHV's on-chip because Intel, et al. were becoming a limiting factor in the goal of the 3D IHVs and not moving it would stagnant their parts at that time.
One thing that’s emerging as transistor budgets increase is that the programmability of solutions are increasing - the idiosyncrasy of discrete units such as
Vertex or
Fragment Processing are quickly dissolving. By DX10/11 we should see this on the PC front, with the transition being complete by the
Next Generation Consoles. So, to say that the presence or not of 'front-end' processing is somehow making the architecture more or less
Emotion Engine-ish is pretty ungrounded.
The ideologies between the
Emotion Engine and what we know of the
Broadband Engine or Cell [Gotta' love the marketing terms] seem pretty isolated to me. Perhaps you can make the case that Cell is more the praxis of massively parallel computing to the theory as seen in the EE - but thats pretty iffy. Correct me if I'm wrong, but I see the Emotion Engine as more the embodiment of Diefendorff's dynamic computing ideal where you have tiny caches, low locality of reference, and are basically playing a fast-paced game of
Hot Potato with your data once it leaves it's external storage.
Cell, IMHO, from what we've seen thus far appears to be an almost polar opposite. As Ben stated, you have a high degree of locality with your data as, in theory, you could have everything on-board in a large pool of eDRAM. From there it's routed to these APUs where they dwell in their own cache and are operated on. Beyond the parallel issue, I see little commonality - and that argument ends abruptly when you realize that
EVERY architecture that deals with modern 3D graphics is nothing but a massive parallel computer - that’s how you do RT 3D, by doing all your ops in parallel.
Having said that, I'd just like to state that I feel nVidia will surpass the PS3 in the tasks historically (heh, looking forward is hard) considered raster functions. I think PS3's ace will be the flexibility and it's
potential for intricate and advanced lighting/illumination that it gives to developers.
Alrighty, it's already 7 and I'm still sitting here... <opens Google>... <search: "obsession"> Chronic lateness kills, damn 'net.