I don't know what results to expect. If I did, I would've simply provided them already.
Do you know anything about science? What's the point in any test if you don't know what results would support or refute your hypothesis?
Were I able to do so, there would be no need for the thread I have created.
So then you have zero basis for suggesting RV670 is texture limited.
A bottleneck that is present 38% of the time is hugely significant, given just how many potential bottlenecks there are in RT graphics rendering. I find your omission of percentages of other bottlenecks to be rather curious, given this fact, and believe your analysis to be disingenuous in this regard.
First of all, I'm saying that 38% is the absolute maximum, and only holds if you assume that the 3870 can do all non-texturing related tasks as fast as the 4850. That just isn't the case, because games where AF has near-zero impact (and hence are probably rarely texture limited) still show huge gains for the 4850. My guess is that it's something like 20% for texturing.
Second of all, I don't have the data to list other percentages. I used only two data points - 3870 and 4850 no-AA scores - and made two assumptions: the 4850 is equal to or faster than the 3870 in the cumulative non-texturing-limited loads, and 2 times faster in the texture limited loads. Actually, I made a mistake last time (I used 2.5x), so the figure I get is <46%. Just to show you how conservative that is, if we assume the 4850 is 20% faster in the total non-texturing-limited load (which is quite reasonable), then the texturing-limited figure drops to 19%.
Anyway, I figured something else out below.
(BTW, stop misquoting me. I never said texturing-limited load is 38%.)
Incorrect. RV770 does not have "faster math speed", it has more math resources working in parallel. This is an important distinction.
No, it isn't important at all. RV770 can crunch through math loads way faster than RV670, hence it has way faster math ability. That's all that matters, especially in the sentence it was used it in.
Pardon? If bottleneck A is a bottleneck for less time than bottleneck B, would that not make bottleneck A less significant?
Remember that I don't share you opinion that bottleneck A (e.g. setup or even BW) is a bottleneck for less time than bottleneck B (texturing).
Feel free to show me another bottleneck that accounts for > 38% and I'll shut my big fat mouth
Like I said, it doesn't have to be 38% (or 46%) to be bigger than texturing because that's an upper bound, but I'll show one anyway.
Per-frame loads that are independent of resolution - which are mostly setup limited - account for 40% of the render time on the 3870 according to the same
hardware.fr graphs (no-AA). Some of that will be shadow map rendering which is always either setup or z-fill limited.
I'm assuming pixels take about the same time to process at both resolutions, which is pretty reasonable considering that there's only 14% more pixels in each direction. With only two resolutions in the data I can't do much else anyway.
It's an IQ enhancer. If you lack AA capability you can simply choose not to use it. The same can't be said for texturing capabilities.
Yes, it can. Texturing ability is also just "an IQ enhancer", because you can simply choose a lower resolution if you lack texturing or shader ability.
R300 was primarily math-bound.
How do you know?
resulting in a chip which was bottlenecked by its math rate less often than the previous chip.
How do you know?
