NVIDIA considers Groq amidst Samsung’s 3nm yield challenges

NVIDIA is introducing a new inference chip based on Groq technology, moving the decoding tasks to Low Latency Processing Units (LPUs) to improve response times. This shift allows GPUs to focus on bulk computations.

Samsung is encountering challenges with low yields in its 3nm and 2nm processes, which may affect production schedules and client confidence in their capabilities.