fablnx/engineeringHow we build Fablnx: GPU kernels, drawing understanding and the measurements behind them, explained from first principles.https://blogs.fablnx.com/Accountable AI starts at the pixel: 2.7–19× faster than NVIDIA’s cuCIM, up to 11× faster than NPPhttps://blogs.fablnx.com/posts/every-colour/https://blogs.fablnx.com/posts/every-colour/How we label every region of an 806-megapixel construction drawing in 49 ms on an NVIDIA A100 (89 ms on an RTX 3070), with labels identical to cc3d's, explained from first principles.Tue, 20 Oct 2026 07:00:00 GMT