Paper 2 (Part B) results: energy to a target price error for Heston, basket and Asian Monte Carlo on CPU, AMD Alveo U50 FPGA and NVIDIA V100 GPU, with co-execution and two-server scheduling

Measured results and analyses of Part B of Paper 2, registered in advance in protocol v0.6, addendum v0.6.1, addendum v0.7 and addendum v0.8 (Zenodo record “Paper 2: registrations”, concept DOI 10.5281/zenodo.23494044). Contents: V100 GPU sessions and checks; stronger frodo CPU baselines and host-wait tests (phase 6); 200-replicate accuracy runs; finer step grid and QE-M sensitivity; CPU+accelerator co-execution (v0.7); a 120-option book scheduled across two servers (v0.8); and the AMD Vitis Libraries 2023.1 engines on the U50. The file RESULTS_PartB.md inside the zip lists every part, where it is, and the outcome of every registered hypothesis. The same files are in GitHub randikat33/u50-fpga-cpu-benchmark, commit 2d269aa835d99e9e90c75b97e1f3c57404e669b2. The results are also in the GitHub repository at commit 706f6c4f8d6982d131778b5f5cb0f6ea7a01888d, with GitHub’s record of every push in paper2/registration_evidence (archived at https://web.archive.org/web/20261030041419/https://api.github.com/repos/randikat33/u50-fpga-cpu-benchmark/activity?per_page=50).