> > But now think that: > > you have 8 light sources (specular, highlight, abient > nicely mixed), > > some 3 to 8 clipper planes, > > an exponentional fog function applied, > > you are using two sided triangles > > and of course misc material sepcifications. > > I suspect that under these conditions, neither does gf3 get > its peak 88 million triangles/second.
It depends on trying that out. As a typical pipleine design will require each stage to perform as fast as the data does run in, these (mostly scenery static) parameters should apply on the fly with only lengthening the pipe pass trough time but not the highly important data troughput rate. > What do you think --- that Nvidia quotes as > the peak performance of the card the performance when drawing the > computation-intensive triangle, or the simplest one? nope, sureley not. But i assume that pretty much features could be applied without any siginficant performance impact. A typical triangle count (=vertex count) over triangle size diagram for current adapters will show a pretty constant vertex rate for small triangles (limited trough AGP bus or pipeline speed), and an 1/x curve for the bigger triangles (limited in fillrate trough rasterizer speed and framebuffer bandwidth). The overall utilisation peak of any current adapter is at the point where the peak vertex rate line meets the fillrate peak curve. Now you turn on the misc features that you want to use in your specific target and see how that diagram changes. But for a true performance value you are best adviced to ignore such academic curves and get hands on a test suite that matches most of the features that your application does use. > Does Conway's Life run faster on GF4 using the stencil buffer > algorithm go > faster than the best implementation on an x86 processor yet? Stencil buffer? Hmm, maybe accum buffer would work as well or better. Call 4 times a blit with accum/stencil, framebuffer clearn and one times a fill that let the board only draws the the pixels which match the respective accum/stencil value. > One site reports that GF2Ultra ran at 16 million cells/second. > However, another reports that a 66MHz PPC was able to do 17 million cells/second. > So maybe we haven't gotten there yet. Further still before a turing > machine implemented in Conway's life running in the stencil buffer > emulating x86 code beats the Pentium 9, I guess. > > Jeff I think a not yet realized "life" rasterizer, texture or blit operation would perform much better. At least the infrastructure in such a grafics chip would serve the purpose much better than if someone wanted to implement the same in a general purpose CPU. Regards, Alex. _______________________________________________ Dri-devel mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/dri-devel