175/365 of GPU Programming
Problem 2 of GPU Mode's Linear Algebra kernels was just released.
This one is about implementing batched real symmetric eigen decomposition on a B200.
Will spend some time today to familiarize myself with the problem and see if I can get a reasonable baseline running against the benchmark shapes.
174/365 of GPU Programming
There's a really beautiful geometric intuition for Householder reflectors/transformations I−2uuH that I'm learning more about today.
I feel like things are starting to sort of click and I'm beginning to see the overall picture. Of course nowhere near expert level understanding yet but it's nice to go from zero knowledge about QR factorizations to being to able to partially comprehend the differences and intuition behind various methodologies thanks to this competition.
Excited to see what the second Linear Algebra kernel challenge is gonna be and what I need to study next!
