Amazon, Artificial Intelligence, Tracy discussed on FT Tech Tonic - The graphics chip powering AI technology

Automatic TRANSCRIPT

So away thing about it is is both are artificial intelligence computing promise both are highperformance computing problems and were using the same architecture for the gogel's in the amazon's to do artificial intelligence and cloud as we are doing artificial intelligence in a car and do you need not much competing halls power ashley in the vehicle all i mean people talking about with five g coming online at largest become have done in the cloud and and this piece of different problems head as the training problem which is how you kind of teach tracy the to think about whatever problem as the try to solve and then this the the sort of deployment which i i guess instead of infants die of a a a you playing and both of those areas and and does every con need the sort of training type chip the architecture for training and influencing is very similar but it's not exactly the same however in our case we are able to take our architecture and adapted for training as wall spree influencing in a car we have to infants and credible speeds and the reason for that is because the late and see that goes to the cloud him back is too long even a hundred milliseconds is too long you know in a hurry milliseconds your car would have travelled hundreds of feet depending on how fast you're going and so you need to make sure that you can inference and you can detect them and reason about what's happening around you and take the appropriate actions instantaneously and so that the latency of going to the cloud and back is too long the predictability of the fact that the you have good enough connection so that you can go to the cloud and back is unpredictable and unguaranteed and so if you were driving in a car you need to have complete confidence that your car is computing all the time how do you make sure that that same kind of technology that powers a big car picked data center wet.

Coming up next