OpenAI's Titan targets inference costs with purpose-built chip, not Nvidia GPUs — Markdown | type0 | type0