Artykuł powstał w związku z tematem Cow3D i renderowaniem licznika.
https://keasigmadelta.com/blog/warp3d-novas-performance-boost-partially-hidden-by-lazy-cowWersja BEZ renderowanego licznika:
https://keasigmadelta.com/assets/Upload ... nfoBar.lhaWersja Z renderowanym licznikiem:
http://aminet.net/dev/src/Cow3D.lha-------- pomocne linki:
1)
http://amigaworld.net/modules/news/arti ... t=40#98720Just discovered that Cow3D's fps counter measures *all* operations, including rendering the line of text at the top containing the fps counter. Text is currently rendered in software, and so involves RAM=>VRAM transfers (CPU-based because it's line segments which are usually too small to use DMA on). This could explain some of the results.
2)
http://amigaworld.net/modules/news/arti ... t=40#98723The length of the command sequences sent to the GPU is longer than you think. However, as I discovered yesterday the fps calculations include the time taken to render the fps counter and info bar at the top of the window (plus whatever else the computer is doing). That increases the amount of PCIe bandwidth used per frame.
3)
http://amigaworld.net/modules/news/arti ... t=40#98731Not necessarily. Cow3D is but one test under specific conditions. Plus, I've discovered that rendering the fps counter & info-bar is skewing the results significantly. Have a look here for my comments and a modified test:
https://keasigmadelta.com/blog/warp3d-n ... y-lazy-cowSo the fps counter does *not* measure purely the Warp3D/Nova performance; it's skewed by the relatively-slow text rendering done afterward (and all other operations).