Leveraging the companies’ joint expertise in the hardware and software market, the collaboration will combine the advanced algorithms and flexibility of PaddlePaddle software and the high-power compute capability of Imagination’s processor cores. The partnership will support chip and application developers in creating fully optimised solutions with greater ease for a range of deep-learning-based implementations.
Featuring Imagination’s heterogeneous GPU, CPU, and AI Semiconductor IP, the partnership offers an efficient and flexible software stack solution achieved through close hardware-software collaboration, ultimately providing design guidance and benchmark references for chip vendors. Imagination and PaddlePaddle have jointly optimised the full-stack solution from model to IP and built a Model Zoo offering reference solutions to platforms and end customers. The partnership will also develop training courses and will offer joint training certification for PaddlePaddle developers who use the Imagination Neural Compute SDK.
Shreyas Derashri, VP of Compute, Imagination, says: “We are delighted to join Baidu, and other partners, in launching the Hardware Ecosystem Co-creation Program to unlock the next level of co-created software-hardware chip design. Imagination has a longstanding relationship with Baidu, developing a joint hardware ecosystem in 2020 and a global AI ecosystem in 2021. Our continued collaboration leverages our joint strengths in advanced computing to deliver more optimised solutions for the industry.”
Qiao Zhao, Head of the PaddlePaddle Product team, Baidu, says, “Imagination, as a leading global processor technology and IP provider, has given great support to PaddlePaddle in terms of both technology and ecosystem. We are excited for our collaboration in various AI applications which further leverages Baidu’s industry-leading deep learning framework algorithms and open platforms, and better supports downstream customers in these markets.”