DEEPX’s ultra low power AI accelerators are designed to optimise on-device processing, ensuring fast, reliable and efficient computation for edge devices. DEEPX technology employs a novel quantisation approach which allows the amount of data handled to be reduced without sacrificing accuracy. More data can thus be processed on the edge at a lower power and a lower cost point.
Many of the real world solutions using DEEPX technology today have been implemented by OEMs who were previously running AI models on GPU-based box solutions. The focus for these users is typically full resolution, high performance video processing running alongside the implementation of AI models – but box solutions can be power hungry. The move to DEEPX technology enables solutions which are lower cost, smaller and less power hungry – opening up applications where AI has previously been dismissed due to power constraints.
Design teams are also implementing DEEPX accelerators into their own board level products on the edge – realising high performance, low power, low cost solutions. We often see brand new applications for which DEEPX AI is proving to be an enabling technology; we believe the real world performance of DEEPX solutions is unrivalled.
Solsta can help you to evaluate the AI accelerator approach and to compare the performance of CPU, GPU and NPU based solutions for your application. Quoted TOPS performance ratings don’t tell the full story, and upgrading the CPU in an application to accommodate the demands of AI is not necessarily the best approach. Pairing a modest processor with an AI accelerator can be a much more efficient way to onboard the power of AI in your application.
Contact the Solsta AI team to explore these approaches and find out how we can help with your AI adoption.