GPUs are becoming key enablers of machine learning, but they are facing competition. Don Clark at The Wall Street Journal reports that Massachusetts General Hospital is planning to use Nvidia chips to spot anomalies in CT scans. The project will draw on a database of 10 billion images.
Clark quotes Keith Dreyer, vice chairman of radiology at Mass General and executive director at the center that is running the project, as saying, “Computers don’t get tired. There is no doubt that this will change the way we practice health care, and it will clearly change it for the better.”
Not everyone is sold on GPUs. Clark reports that Google disclosed last week that, in addition to Nvidia GPUs, it is using an internally developed processor for machine learning. Other companies working on GPU alternatives include Movidius and Nervana Systems. And IBM’s TrueNorth chip has an architecture inspired by the neurons, synapses, and other features of the brain.
Clark quotes Jeff Hawkins, cofounder of Numenta, which works on brain-like forms of computing, as saying, “There is no way that existing [chip] architectures will be right in the long term.”
Nevertheless, Nvidia’s lead in the field has caused its stock value to double in 12 months, pushing its market value above $24 billion. Clark writes that the company’s data-center business has risen 62% from a year earlier.
Clark quotes Nvidia CEO Jen-Hsun Huang as saying, “It is now clear that hyperscale companies all around the world are moving into production.” He also cites Tractica LLC estimates that spending on GPUs for deep learning will grow from $43.6 million in 2015 to $4.1 billion by 2024, and related software spending will increase from $109 million to $10.4 billion over the same period.
Regarding Google’s work on a GPU alternative, Huang said an earlier generation GPU was better at training than making analytical decisions once the training is completed, but the new GPUs are much faster at the latter task.
Medical applications are far from the only applications for GPUs and related devices. Blue River Technology has trained tractor systems to decide where to spray herbicide. Clark quotes Ben Chostner, vice president of business development, as saying, “Those machines are making 5,000 see-and-spray decisions a minute.”
