OpenAI and Broadcom unveil LLM-optimized inference chip
DGX agentOpenAI and Broadcom announced a collaboration on a specialized inference chip designed to optimize the performance and efficiency of large language model deployments. The chip, referred to as 'Jalapen