Cerebras gives waferscale chips an inferencing twistCerebras' new inference accelerator uses SRAM for superior performance in generative AI, highlighting the shift from memory bandwidth limitations to processing capabilities.