AMD and Cerebras partner for AI inference, claiming efficiency gains over Nvidia's SRAM decode approach | HyperFast