Kyle Yu

Yu’s Red Hat internship involved Triton kernels and parallel programming. His 2025 DevConf session introduced kernel-writing pitfalls and work on BLAS routines. After completing his Boston University degree, he announced a role at Tensordyne working on infrastructure for an inference chip that uses logarithmic arithmetic.[5][1]

Connect a thread’s address to the memory work it creates

BrrrViz uses interactive chapters to explain execution and memory behavior. Its coalescing lesson shows how access patterns change the GPU’s loading work; its divergence lesson shows a warp following different execution paths. Another chapter explains shared-memory banks and replay when threads access different addresses in the same bank. The controls let a learner change a pattern and inspect the consequence, rather than only memorize the terminology.[6]

Show a kernel’s failure as well as its fix

Triton Kernel Misadventures collects examples of incorrect global reductions, atomic-operation accumulation and multiple blocks writing the same location. Its stated goal is to connect each failure with an explanation and a corrected approach, with results checked against reference implementations.[7]