Your browser is not supported. Please update it.

12 February 2016

Imagination faces second year of losses as GPU market toughens up

The mobile GPU (graphics processor unit) market is a tough one. The main GPU cores in the space – Imagination’s PowerVR, ARM’s Mali and Qualcomm’s Adreno – are suffering from the slowdown in smartphone growth just like other components. There are significant new opportunities for firms which can set the pace in emerging GPU technologies, especially those that will power new graphically intense user experiences based around virtual reality. But that requires significant R&D investment, so many players, even the high end groundbreaker Nvidia, will face a difficult hiatus between the old and new growth engines.

Imagination Technologies is showing the strain. Its CEO, Sir Hossein Yassaie, who has run the company since 1998, has stepped down as the IP licensing firm heads to a second year of losses. It now plans to sell its digital audio unit, Pure, which has been a persistent drag on its results, but aims to invest more in keeping its PowerVR core designs at the forefront of the market, especially as their most prominent customer is Apple. Imagination said it will cut around £15m from operating expenses in its next fiscal year, but will reinvest around £2m of that in PowerVR.

Imagination also owns CPU core provider MIPs, but in both its GPU and CPU businesses it has been under intense pressure from larger UK rival ARM. MIPS has largely failed to make inroads into ARM’s domination of the mobile processor space, though it is showing some signs of leadership in wearables and has a valuable foothold among Chinese IoT chip firms. Meanwhile, Apple’s Mali, for years a definite underdog to PowerVR, overtook its competitor in annual shipments in 2014.

Imagination reported a £13m ($19m) loss on revenue of £177m in its last fiscal year, which ended on April 30 2015, and is likely to make a second loss for fiscal 2016. When it reported half-year figures for the period to October 31, it forecast a better second half would return it to profit for the full year, but it now acknowledges that has failed to happen and it expects a pre-tax loss.

It said two factors were to blame – lower than expected numbers of new licensing deals for its cores, and lower than expected royalties from some existing licences with key customers. The latter was widely interpreted to mean Apple, which has seen a fall in iPad sales and flat growth in iPhones over the past year.

But like all companies which rely heavily on smartphones, the quest will be on to expand the platform into new sectors and push the technological boundaries. GPUs are a hot area of blue-sky research as companies and universities try to define the next generation web experience and user interface, and some of their current findings will be feeding their way into commercial companies in the coming years.

For example, researchers at the Massachusetts Institute of Technology (MIT) have developed an experimental GPU called Eyeriss, which could enable algorithms to run locally and instantly, without sending data to the cloud. That would make new ways of searching the web far more responsive. Technologies like Apple Siri and Microsoft Cortana are early indicators of the move to harness machine learning and AI in the cloud, in order to deliver highly personalized, context aware answers to users’ queries. With Eyeriss, the neural network which pulls all the data together to answer the question could live on the mobile device, not the server.

The mobile GPU design, presented at the International Solid State Circuits conference last month, claims to be 10 times more efficient than current mobile GPUs and so capable of running AI algorithms locally without killing the device’s battery – which would also be valuable in situations where connectivity is not always available. The MIT team envisage several uses for in-device neural nets, such as battery powered autonomous robots, and various IoT appliances.

Other teams are closer to the real market with systems that embed AI algorithms and vision processing on a device. Qualcomm has combined its Zeroth cognitive capability with Snapdragon to offer this capability in cars, also a target for Nvidia’s work, as well as robots and smartphones. IBM has been adapting some of its Watson algorithms for client devices while Google recently announced a partnership with embedded vision company Movidius for on-device machine learning, which it could apply to driverless cars and mobile gadgets.