LLM infrastructure cost reduction via NUMA-aware weight banking: 147 t/s (8.8x stock llama.cpp) on refurbished enterprise POWER8. Self-hosted inference, no cloud APIs. Part of the Proof of Physical AI stack.

self-hosted transformer numa powerpc energy-efficiency ppc64le cost-optimization power8 inference-optimization edge-ai hebbian depin ai-inference llm llmops llama-cpp llm-inference ai-infrastructure proof-of-physical-ai vec-perm

• Updated Nov 7, 2026
• Python

llama.cpp optimizations for IBM POWER8: vec_perm non-bijunctive collapse, PSE hardware entropy, DCBT prefetch. Sovereign inference. Part of the Proof of Physical AI stack.

transformer ibm powerpc pse ppc64le power8 ai-inference llm llama-cpp llm-inference vintage-computing sovereign-ai vintage-hardware sovereign-inference proof-of-physical-ai vec-perm

• Updated Sep 26, 2026
• C