Herdora Headlines

Wafer Blog

This article from Wafer's blog details a benchmark blind spot encountered during GPU kernel evaluation, where JIT compilation and CPU core migration can skew performance measurements, and provides a fix for accurate benchmarking.

Companies:Herdora

Fondo

Herdora launched 'Cursor for CUDA', a tool that automatically optimizes PyTorch GPU code, aiming to democratize GPU performance for AI applications. It addresses the challenge of slow ML inference by generating optimized CUDA kernels and monitoring production models.

Companies:Herdora