plan.md

Experiment 03: Memory Copy Baseline

1. Focus

• Characterize simple read, write, and copy kernels as a practical memory-throughput baseline.

2. Question

• How much bandwidth does the GPU sustain for the simplest contiguous memory modes?

3. Variants

• read_only
• write_only
• read_write_copy

4. Method

• Run one simple contiguous kernel per memory mode across a size sweep.
• Time dispatch separately from host transfers and validate each mode against a deterministic reference.

5. Outputs

• Median GPU time by mode and size.
• Effective GB/s by mode.
• Saturation knee and steady-state comparison across modes.

6. Interpretation

• This is the closest thing to a roofline-style denominator for later bandwidth claims.
• More complex kernels should be compared against this baseline before claiming efficiency.