https://msoexpert.com/ Local AI inference, MSO infrastructure, telecom architecture, and enterprise technology from John Pezzulli. en-us Sun, 09 Aug 2026 16:00:00 GMT https://msoexpert.com/articles/the-model-is-the-silicon/ https://msoexpert.com/articles/the-model-is-the-silicon/ Sat, 08 Aug 2026 16:33:03 GMT Taalas physically embodied Llama 3.1 8B in silicon, reached roughly 17,000 tokens per second for one user, and was acquired by AMD before I finished writing about it. https://msoexpert.com/articles/amd-ai-pro-r9700-rocm-vulkan/ https://msoexpert.com/articles/amd-ai-pro-r9700-rocm-vulkan/ Tue, 04 Aug 2026 22:11:07 GMT ROCm worked on the first try. Vulkan was substantially faster, and the hardest failures belonged to everything around the card. https://msoexpert.com/articles/making-deepseek-v4-flash-fit-one-96gb-gpu/ https://msoexpert.com/articles/making-deepseek-v4-flash-fit-one-96gb-gpu/ Tue, 04 Aug 2026 00:47:58 GMT Five mapped expert layers made room for speculative decoding, a correction tier, and near-million-token context. https://msoexpert.com/articles/living-room-ai-edge-node/ https://msoexpert.com/articles/living-room-ai-edge-node/ Mon, 03 Aug 2026 17:08:34 GMT The origin story, hardware escalation, runtime experiments, failures, and corrections behind a private local AI system.