Fournex Blog
GPU Performance, Explained
Practical guides on CUDA profiling, bottleneck diagnosis, and optimization techniques — written for engineers, not slide decks.
CUDAProfilingEvaluation
Can a Profiler Judge AI-Generated CUDA Kernels Without Cheating?
We pointed Fournex at SakanaAI's archive of AI-generated CUDA kernels to ask whether our analyzer can produce useful, evidence-backed judgments without ever seeing a human label.
June 17, 20269 min read
Read articleCUDAProfilingPerformance
Why Is My GPU Idle? A CUDA Bottleneck Checklist for Engineers
"GPU utilization is low" is not a diagnosis. Walk through the five root causes — host-device stalls, uncoalesced memory, low occupancy, sync overhead, and tiny kernels — and learn exactly what to look for in Nsight Compute.
June 14, 20268 min read
Read article