By: VertexMaster (nope.delete@this.nope.com), August 5, 2016 5:16 am
Room: Moderated Discussions
> Look closer. It is the same 8x8 tile (actually 4x4 quads).
>
> Tiles seems to be placed in memory sequentially in case of David's card.
*shrug* it's still scanline rasterization. Commercial software-based scanline rasterizers have been doing that type of "block" rendering since the 90s with SIMD.
Getting back to GCN 1.2, it looks a lot like Maxwell's output. It's block rendering for the benefit of cache/compression. We can call it "tiling", but it's nothing like TBDR and more importantly, not some key advantage that Maxwell has over Kepler or AMD's GCN 1.2+ cards.
http://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/pubs/doggett12-tc.pdf - Covers caching in GCN 1.0 supporting the idea it's just for cache coherency.
>
> Tiles seems to be placed in memory sequentially in case of David's card.
*shrug* it's still scanline rasterization. Commercial software-based scanline rasterizers have been doing that type of "block" rendering since the 90s with SIMD.
Getting back to GCN 1.2, it looks a lot like Maxwell's output. It's block rendering for the benefit of cache/compression. We can call it "tiling", but it's nothing like TBDR and more importantly, not some key advantage that Maxwell has over Kepler or AMD's GCN 1.2+ cards.
http://fileadmin.cs.lth.se/cs/Personal/Michael_Doggett/pubs/doggett12-tc.pdf - Covers caching in GCN 1.0 supporting the idea it's just for cache coherency.
