fulcizombie said:
Bottom line:I believe that all this talk about "efficiency" is bullshit excuses and PR "smoke and mirrors" and will look moronic if the first ps3 games look much better than the xbox360 launch games(which will probably be the case).
First PS3 games are entirely... guaranteed to look better. PC and Xbox ports with development... no, tweaking, for a few months on actual architecturally-accurate chips. Versus the right architecture for well over a year. The comparison becomes completely and utterly useless as a hardware benchmark, and this should be absolutely and utterly obvious. So it'll look moronic. Nobody's fault except the individuals if they fail to account for the other factors, so from the hardware (analysis) perspective it shouldn't matter if it looks pathetic in a comparison of launch title visual quality.
And in the real world, efficiency is absolutely and utterly important. Raw power means squat when
usable energy is the product of the raw, theoretical value and your efficiency. What's the point if you can do two multiply-adds per clock in a pipeline, tops, if you get one cut down due to a texfetch or because you can't coissue the instructions you want for that clock cycle? Zero. You have, realworld, one in that case.
Alot of the things these companies state aren't invalid in of themselves. It's that people fail to put them into proper context. A Gflops number isn't useless. You just need to know how they're trying to BS you and avoid it. It's great you have X number of Gflops. Whether I know that you lose Y because of normal, standard rendering tasks that are taken care of by hardware that is shared amongst numerous tasks, or lose K because it's really only there for special cases is my problem.
Nicked said:
Sony just do it well, thats a good thing.
Please, tell me that I am misinterpreting your statement, as the way I read it now, it's a good thing for a company to attempt to mislead me. Don't think I actually WANT to wade through all the extra BS. But perhaps I'm alone in that.
tema said:
I heard that Xenos is closer to R480 in perf. The line is blurred between exotic_unproven, unified_competing.
The only time I've ever heard R480 referenced when speaking of Xenos was in a flawed comparison between the two, in which the conclusion was that it would act like a 20-pipe R480 (or something like that). A lovely comparison that figured a pipeline that occasionally coissues an add before the full-ALU op is worthy of being considered {edit: just as] capable as using a completely capable ALU unit in two subsequent clocks. As stated by ATI, however, Xenos will surpass R520 at the targetted resolution (whilst losing out at higher resolution, which implies hitting fillrate or having to reprocess geometry for probably >half a dozen tiles, as would be the case for AA'd 1080p framebuffer), which should be a satisfactory cut into that comparison for you.
So long as shaders keep to their present math: tex ratio, Xenos can remain quite limited, though. I don't think X1600 quite gets the 3:1 math advantage it holds over R520, but it certainly doesn't perform like a 4 pipe card either. It gets a little extra whenever it can issue an op for the add/imput modify-only ALU before another ALU op. Having seen those two code snippets posted at Rage3d recently, that mini-ALU doesn't seem to have many situations where it actually comes into play, and math to tex ops are already pretty high. Which really begs the question of why G70 doesn't perform better, since it should have considerably greater opportunity to use that second ALU. Or perhaps I'm too tired to think straight.
But, of course you've taken to those "exotic vs. unproven" and "unified vs. competing" lines. Where did they come from again? They're fairly brilliant statements, worthy of some admiration. IIRC, what's stated is "it takes careful work to balance between the work you need done and what's being done" and the result is people take it as "it's pretty likely there's going to be serious issuuuueees!" That's brilliant PR BS at work, right there. The best gives you a simple fact and lets you put the spin they want on it. They haven't lied, but they've still gotten the same message aross that they wanted to.
