Project Dominion Studios · Whitepaper 01
The Pascal Frontier
Modern AI inference on hardware that apparently didn’t get the retirement memo.
A graphic-novel technical field report about qualifying vLLM and llama.cpp on Lurch’s NVIDIA Tesla V100—successful paths, stubborn failures, measured results, and receipts included.
- V100 · 32 GB
- vLLM vs. llama.cpp
- August 2026
Translated for family, friends, leaders, and curious humans who did not voluntarily memorize CUDA compatibility tables.
01
The question
What can a 32 GB Volta accelerator still do when current software support begins treating it like yesterday’s hardware?
02
The engineering
Build, test, break, measure, compare, and document the paths that actually work across vLLM and llama.cpp.
03
The result
A public, reproducible account of capability, compatibility boundaries, thermal behavior, and real-world tradeoffs.
Evidence before hype.
The publication separates observed results, analysis, conclusions, hypotheses, and planned work. Supporting runs, metrics, scripts, claim mappings, and redaction records live with the public release.
Scott directed the research and approved the publication. AI systems assisted with exploration, drafting, analysis, production, and QA; they did not autonomously approve claims or release.
Read the evidence-led publication →