Imagination Technologies Secures $100M Investment for AI Reboot
Imagination's GPU designs have been used by a range electronics manufacturers over the years. Photo by Chris Ried on Unsplash.
Key Takeaways
- Imagination Technologies has secured a $100 million convertible term loan from Fortress Investment Group.
- The semiconductor designer is in the middle of overhauling its AI offering.
- After seven years, the firm has stopped developing dedicated neural network accelerators to focus on GPUs.
The embattled UK-based computer chip designer Imagination Technologies has secured a $100 million convertible term loan from Fortress Investment Group as the company attempts to “reboot” its struggling business, CEO Simon Beresford-Wylie said on Friday, Aug. 26.
Significantly, the company has ditched its neural processor range to focus on the GPU market and expanding its software offering.
Chipmaker Backs GPUs Over NPUs for AI Processing
Although Imagination is most known for its GPU patents, in 2017, it branched out into neural processor units (NPUs)—dedicated hardware for neural network training and inference.
“Like pretty much every other company, we saw convolutional neural network accelerators and thought, ‘Yeah, we should do one of those,’” Chief Innovation Officer Tim Mamtora said in an interview with EE Times.
After seven years, however, the company has reversed course, abandoning the NPU project to focus on optimizing its GPU designs for AI workloads.
The decision reflects the innate challenges of rolling out a new technology like NPUs. Although they are more efficient than GPUs, the software stack for NPUs is much less developed. Ultimately, equipping developers with the necessary tools to use the new chips wasn’t worth the extra hassle for Imagination.
As Mamtora explained:
“It became a huge software task, which morphed from hardware we could develop in-house to having to have a big team of software engineers [with] customers needing so many levels of support. We didn’t see that as sustainable.”
Of course, doubling down on GPUs means Imagination has to find space in a market dominated by Nvidia.
Catching up With Nvidia
While both companies operate in the GPU space, they aren’t direct rivals in most markets. While Imagination’s traditional strength lies in edge computing, with a focus on mobile chips and embedded solutions, Nvidia’s success in recent years rests on its AI data center offering.
Different business models also distinguish the two companies. In contrast with Nvidia’s full-stack hardware and software offering, Imagination typically licenses its designs for other companies to use.
But as the firm repositions itself in the AI chip market, Imagination is expanding its range of solutions for data centers and building out its software offering to compete with Nvidia.
Having terminated its NPU initiative, the company has thrown its weight behind the open-source AI development platform SYCL, which Mamtora said he envisaged becoming an alternative to Nvidia’s CUDA.
Developing an Alternative to CUDA
The popularity of the CUDA framework, which is used to run applications on GPUs, has helped Nvidia assert its position as the GPU titan it is today. What’s more, because crucial machine learning libraries like PyTorch are optimized for CUDA environments, the platform has become integral to modern AI development, helping cement the industry’s reliance on Nvidia chips.
Incidentally, the near-ubiquity of CUDA within AI development is part of the reason NPUs and other GPU alternatives haven’t taken off despite offering significant performance and efficiency gains.
In contrast, SYCL is a chip-agnostic framework that is compatible with a variety of semiconductor architectures.
South Korea’s Latest Move to Dominate AI Chip Market, a New Nvidia?
US Chipmaker Microchip Suffers Cyberattack in Latest Semiconductor Targeting
