Orca Family

Orca’s ability to process pruned models essentially multiplies compute and bandwidth resources, allowing for the processing of larger, more accurate models to enable real-time, cloud-like quality applications at the edge.

Having DRAM access support in our AI accelerator offers significant advantages. By enabling direct DRAM access, our accelerator provides faster data transfer rates, resulting in improved performance and reduced latency. This feature also allows for quick, seamless switching between neural network (NN) models, minimizing downtime and boosting productivity. For applications like image or speech recognition that require frequent model changes, this capability adds flexibility and efficiency, allowing users to easily handle varying datasets or tasks.

At DeGirum, we take pride in our AI accelerator’s flexible architecture, supporting both int8 and float32 precision formats. This versatility allows customers to optimize performance, accuracy, and power consumption based on their specific use case. Additionally, our AI accelerator features intelligent power management, dynamically allocating resources according to workload, minimizing power usage without compromising performance. With its adaptable architecture and smart power management, our AI accelerator provides a scalable, efficient platform for a wide range of AI applications—whether handling large datasets or requiring real-time processing, it delivers exceptional performance, accuracy, and energy efficiency.

Our AI accelerator’s highly scalable architecture makes it an ideal solution for organizations of all sizes. Adding additional boards with our chip results in a linear performance increase, offering a seamless upgrade path as your needs grow. This scalability allows for easy, cost-effective expansion of AI infrastructure, while ensuring smooth integration with existing systems. In response to customer demand, we can develop boards with multiple chips, leveraging our chip-to-chip interface for even greater scalability. This flexibility empowers customers to build larger, more complex AI systems with ease, while maintaining the linear performance scaling that sets our AI accelerator apart.