Thien Tran(@gaunernst):Thank you @pranjalssh for the shoutout 🥰 Some thoughts: - Reading lane_id is interesting... Usually I do warp_uniform() on warp_id (broadcast from lane0) + elect.sync. Wonder if there's perf advantage (at the mercy of ptxas) - https://t.co/ZYtiK2HPzh.L1::no

2026.08.17 05:39

Thank you @pranjalssh for the shoutout 🥰 Some thoughts: - Reading lane_id is interesting... Usually I do warp_uniform() on warp_id (broadcast from lane0) + elect.sync. Wonder if there's perf advantage (at the mercy of ptxas) - ftw

Pranjal@pranjalssh

2026.08.15 18:26

Sequel for my H100 blog is out now. We implement Blackwell matmul for NVFP4 from scratch, and outperform cuBLAS by 4.7% at N=8192. This time we have something better than Hilbert curves - that exact-fits for the Nvidia hardware. Full blog post with all details: