New paper: "Software Rasterization of 2 Billion Points in Real Time"
- Adaptive precision to better utilize memory bandwidth
- Now with support for frustum culling and LOD rendering
- Throughput: 144 billion points/s
paper: cg.tuwien.ac.at/research/publi
source: github.com/m-schuetz/comp
Conversation
Replying to
That's a lot of points! Can an you write another paper at 3 Billion, or does it have to be orders of magnitude?
1
1
haha yeah, standard C++ sucks quite a bit without some essential helper functions. 😅
This *does* use a GPU though, right? When I saw "software rasterization" I assumed CPU only.
1
3
Replying to
Wow looking great as always, will have a look! What's the vram usage for rendering the 2B?
1
1
Replying to
2B is actually the real-time throughput. Largest we've fit on the GPU is 1 billion points (16GB) rendered in 6.4ms, so we can easily render 1 billion points twice in real-time. We should've stuck with the working title: "Software Rasterization of a Shitton of points in Real-Time"
10