I Set Out to Learn NRVO and Ended Up Measuring My Command Line

I wanted to understand named return value optimisation, got curious what the pessimised version costs, and my first numbers were reproducible to 0.01% and wrong. Instructions and cycles instead of wall-clock, and the pointer question turned out to be an allocator question.

How I Benchmark on the GB10 CPU

The same benchmark landed at 3.47 ms and 7.01 ms on different afternoons while reporting under 0.3% variation both times. These are the habits that survived that.

NVLink-C2C Removes the Copy, Not the Synchronization

The CUDA Programming Guide assumes the host and the device keep separate memories with a copy between them. My DGX Spark doesn’t. So I went and learned what NVLink-C2C actually is compared to PCIe, how coherence behaves across it, and what that changes for someone writing C++ on this machine.