AMD acquires AI chip startup Taalas to boost inference performance by etching models into silicon

AMD has acquired the AI chip startup Taalas to enhance inference performance. The company utilizes a method of etching models directly into silicon to create model-specific integrated circuits. Early technical demonstrations have showcased the capabilities of this technology, specifically featuring Llama 3.1 8B. These demos indicate that the integrated circuits can achieve processing speeds of up to 17,000 tokens per second, significantly boosting the efficiency of AI model execution through specialized hardware.