Imagination Technologies, best known today for the GPUs used in Apple's iPhones and iPads, has made a big leap in its AI strategy by dropping its dedicated NPUs and adding similar capabilities to its GPUs, according to EE Times. Coincidentally, the company just raised $100 million in funding from Fortress Investment Group.
Imagination has accomplished a lot in the 40 or so years since it was founded. Still, the company became known for its PowerVR Kyro-branded graphics processors for PCs and its PowerVR-badged GPU IP, which was used in GPUs from Apple, Intel (and dozens of others). Imagination is also a player in the AI space; the company has “rebooted” its approach to AI by discontinuing the production of standalone neural network accelerators and focusing on building AI capabilities into its GPU IP products.
In an interview with EE Times, Imagination acknowledged that the shift in strategy over the past 18 months was due to the complexity of its proprietary software stack not keeping up with the rapidly evolving diverse needs of its AI customers, which required a custom solution.
Due to Nvidia's CUDA dominance, the challenges didn't end with hardware development. Creating an AI software stack that could expose the capabilities of ImgTec's NPUs to developers and customers proved to be a major hurdle. Customers often submitted their own AI models and expected Imagination to optimize them for their hardware. This created intense competition and pressure for Imagination's development team, which was feeling stressed without any financial results.
Software is the key.
Recognizing these challenges, Imagination shifted its focus to GPUs, which are naturally multi-threaded and naturally well-suited for tasks that require efficient parallel processing and data movement. The company believes that the flexibility of GPUs, which can be enhanced with additional AI-specific computing capabilities, makes them ideal for edge AI applications, particularly on devices with existing GPUs, such as smartphones, ImgTec's specialty.
To support this transition, Imagination is repurposing technology developed for its now-discontinued accelerators, such as graph compilers, and integrating them into its GPU stack, allowing the company to competitively serve AI workloads while leveraging its strengths.
For example, Imagination worked with the UXL Foundation on SYCL, a framework designed to compete with Nvidia's CUDA, a change that also meets the needs of Imagination's customers who already use on-chip GPUs for AI and graphics processing.
Perhaps it was that openness that helped Imagination secure a $100 million convertible long-term loan from an affiliate of Fortress Investment Group.
According to EE Times, Imagination is committed to its GPU-focused approach, but is open to reconsidering dedicated AI accelerators in the future as AI software infrastructure evolves. For now, the company believes compute-oriented GPUs are the best platform to meet its customers' AI needs.
