Integration of AI features is growing, says Imagination
The launch of Imagination’s latest GPU, the DXTP GPU IP, reflects how this solution is responding to an increased integration of artificial intelligence (AI) features into markets like smartphone and automotive, Matthew Bubis, Director of Product Management, told Electronic Specifier in a recent briefing.
Imagination creates GPUs for several markets, smartphone and automotive included. The smartphone market reportedly experienced a slump of shipments from 2017 all the way to 2023, but figures from Counterpoint Research suggest that this will turn around, which partly can be attributed to the integration of AI features in phones.
The addition of these AI features ultimately cannot impact the design of the smartphone, such as its size or its power consumption, as consumers won’t be willing to give up these functionalities. As Bubis put it: “We’re adding all of these AI experiences into the phone, but [the] form fundamentally has to stay the same.”
For Imagination’s customers, OEMs who are looking to integrate these AI features, a concrete example of how this could be achieved includes large language models (LLMs) supporting chat-based AI models.
“The desire would be to have chat-based AI models running locally on your phone, rather than being connected to the Internet … having a long response time,” explained Bubis. “It might be that apps and gaming have a smaller language model embedded in the device, running in the background, to support what the app might want to do.”
In creating GPUs for the smartphone market, Bubis told Electronic Specifier, it lays the “foundation” for solutions that can be applied across other verticals, such as automotive.
“Making sure that it’s energy efficient and drives performance density, making sure our GPUs are as small as possible … the smartphone market is often the technological endpoint which pushes it furthest forward,” said Bubis. “But then we take that fundamental architecture and it still has value across all of our markets.”
According to Bubis, “the benefits we’ve gained through the smartphone market eventually find their way into the automotive market but they tend to be a few years behind”.
The DXTP GPU IP
The new DXTP GPU IP is the latest in a long line of series from Imagination, spanning the A, B, C and now the D series. From A to B, the company focused on improving power efficiency, from B to C the performance was pushed up, and in the D series this became about performance density.
Notably, the new GPU delivers 20% more power efficiency compared with its DXT equivalent. The architecture supports both AI and machine learning, compute and graphics to run at the same time.
Fundamental architecture changes in the D series have resulted in being able to provide more bandwidth for geometry and compute related workloads, to prevent memory bandwidth constraints. “By opening up the memory and cache bandwidth of the GPU, we’re allowing that higher sustained performance,” explained Bubis.
Being able to manage these different workloads means, for example, if two people are in a car, one can play a game in the back while the other can map a route, and it won’t affect the performance.
“Increasing the memory and local bandwidth is good for energy and performance efficiency, because it keeps all of that computing on the GPU chip,” added Bubis. “That’s important, because if all of the computing is passing data to other memory devices, that causes the performance to drop off.”
In focusing on flexibility for developers to be able to program the hardware easily, Imagination is working with the Google Android Ecosystem on developing TensorFlow Lite, Google’s performance runtime for on-device AI, effectively meaning that developers can build their networks on this framework.
“This is about ease of use, flexibility, allowing software developers to easily access all of the compute capability that we provide with the hardware,” Bubis summarised.
AI in automotive
The use cases for the GPU in automotive include vehicle infotainment systems which provide the driver with planning routes, music streaming, heating and cooling settings, and more. “The user interfaces inside of cars are becoming modern,” shared Dennis Laudick, Vice President of Product Management. “They’re driven by our experiences and our smartphones, and we expect them to be similar to that.”
One example of this can be observed in how more modern cars are being designed with built-in screens, which besides having a sophisticated display behind it, are powered by GPUs for smooth graphics processing.
This is particularly noticeable in Asian countries, which Laudick attributed to a “byproduct of the culture”, more specifically an openness to new technologies, but said he expected this to permeate Western countries eventually.
“The automotive industry is one of the first adopters of AI within the GPU and the things they’re doing is revolutionising everything,” Laudick added, “from being able to interface the car to driver detection: is the driver drowsy? Is the driver paying attention?”
This is pronounced in autonomous vehicles, which are reliant upon technologies like AI interpreting their surroundings and being capable to map paths and make decisions in real time. One of Imagination’s customers, Zelos, has already integrated the new GPU into their autonomous vehicles.
In finding the foundations in designing GPUs for the smartphone market, Imagination has identified a formula for success, creating solutions that benefit across multiple different industry verticals.