Technology Introduction:
Bertha is a 4nm AI prcocessor with LPDDR5x, designed for efficient Generative AI inference. Since infernece is bottlenecked by memory, Bertha has streamlined memory access and LLM-optimized compute dataflow to achieve 90% bandwidth utilization compared to GPU's 50%. It features highly scalable chip-to-chip networking to accelerate models of any scale. Bertha delivers 2x higher performance, 3x lower power consumption, and 10x better cost than GPUs.
Industry Applicability:
Bertha significantly reduces the total cost of ownership of cloud service providers when running Generative AI services and encourages wider adoption of AI in more diverse industries. Bertha is expected to replace the expensive GPUs from the datacenters for greater sustainability. Since Bertha is scalable, we also apply Bertha in edge computing to enable personalized AI in IoT devices and home appliances.