![]() ![]() It's a weird memory leak that I haven't been able to reproduce to this extent, but I do notice an increase in RAM usage over time. I do not recall the exact situation but perhaps the game had been running for a few hours by that point in time. I have also noticed that at some point my RAM usage was through the roof. I have stopped playing DS3 after their server maintenance as I finished it, however, I might soon try the game again and see if an update solved the issue or if it's still something with my machine. I can see that in the entire hour of your stream you have not had any type of degradation, which is weird. It is still far from perfect but most of the time it is in a range 30-40 fps The good part is that I haven't experience degradation that you have described. The frame rate got more stable + start using experimental native build of obs to do recording. ![]() But if VRAM pressure is high, the temporary buffers will likely be purged in order to make room, which will again require us to run the vertex and tessellation control stages, with everything that FYI: I did the same setup as you did with d3d11 and dxgi. control stage that way, we pay the performance cost only once instead of every draw. The one thing I haven't tried that might help is caching the output of the tess. I've already tried several things to speed up tessellation on Metal, to no avail. There is unfortunately little I can do about this. Much of the performance gain of TBDR is due to deferring this expensive flush to VRAM the tessellation programming model in Metal, however, completely negates this. On IMR GPUs, this isn't so bad, because they render directly to VRAM, but it's especially bad on TBDR GPUs like M1 because they render to an internal tile buffer, which is flushed every time the render pass ends.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |