Please note any errors in the R520 article here:
use of a new extension to DirectX know[n] as Render to Vertex Buffer.
Shader ALU's by breaking the tasks down into smaller, chunks that can be interleaved more effectively.
a more optimal method of High Dynamic Range blending, something that ATI's part have not been able to to previously.
ATI's engineers state that they have place[d] a lot of emphasis on being able to run with high quality options on.
in fact, RV515 went relatively smoothly, with no unexpected issues, so it is one of the chips that to primarily driving the technology announcements and release on.
You're talking about Dave and his ambitions regarding... what?Mintmaster said:BTW, it's great to see that you're striving for perfection!
Mintmaster said:BTW, it's great to see that you're striving for perfection!
For the parts in the R5xx series that have three times the number of shader pipelines to ROP's the batch sizes are 48 (4x12) pixels large, so the efficiency drops a little.
passing the data between the boards via the chipset does have performance, limitations especially when the chipset uses a single x16 lane implementation split into two x8 configurations
Note: Adaptive AA has presnetly been qualified by ATI on the Radeon X1000 series, however all boards from 9500 up to X850 have the capabilities for performance Adative AA and ATI state that they are going to go back and qualify the feature for use on all these products.
PowerVR's VillageMark benchmark, which draws a scene with very high overdraw levels, does display a fairly considerable improvement on X1800 XL, but exactly how much of this difference is due to the different vertex rates is difficulat to say.
For the tests without branching we see that the X[1]800 XL is performing very close to it theoretical difference to the X800 XT in the tests, even exceeding it in the VS2.0 test,
Our Doom 3 "Turkey Baster" test uses and extended sequence from the Alpha Labs 1
so this makes the capability of supporting branching in the Pixel Shader much greater - how much this will spur developer on to utilise the capability will have to be seen yet.
X1800 XT % Faster than X800 XT
In our Far Cry test we see that using both of these IQ features together, naturally, produces a greater performance hit, however the reduction is greater than the sum of enabling the features individually. In this case we are getting a greater performance hit sometimes because the transparencies are occurring on textures which are now being over sampled for FSAA purposes and they are also having 16x filtering applied to them as well, some of which would have had under the Angle Dependant filtering.
In our Far Cry test we see that using both of these IQ features together naturally produces a greater performance hit than either individually. However, in some cases the reduction is greater than the sum of enabling the features individually. We see this because the transparencies are occurring on textures which are now both being over sampled for Adaptive FSAA purposes and having extra 16x filtering applied to them (some of which would have been reduced under the Angle Dependent filtering).
Here' we'll....
"The memory layout of the XL board is against very similar to the XT"
instead of the second (red colored) "gains" IMO "improvements" sounds a bit better.Comparing the gains the the X1800 XL has in relation to the X800 XT again highlights the efficiency gains R520 has over its previous generations,
you probably meant "faster" oder "higher" here, and the second "being" looks superfluous to me.Curiously, the result at 1600x1200 for the X1800 XT here is actually slower than at 4x FSAA, despite there being more FSAA and texture samples being calculated.
should be the case => should be the opposite (?)When taking all of these performances into account we can see that 4x FSAA certainly has improved with these new memory adjustments, but what sticks out is that 6x FSAA on the XT is actually faster than 4x was in the first place, which logically should be the case as 6x FSAA will always be requiring more more samples to be calculated and more bandwidth used - ...