IoT drives computing on the edge
Nadella has previously spoken of the company’s vision as cloud first, mobile first, but this is now changing with new IoT workloads. He described IoT as a cloud workload that can generate vast amounts of data, and calls for a rethink in how applications are engineered.

“The platform shift is all about data. When you have an autonomous car generating 100GB of data at the edge, the AI will need to be more distributed. People will do training in the cloud and deploy on the edge – you need a new set of abstractions to span both the edge and the cloud,” he said.

This leads to a change in Microsoft’s manta. “We are moving from mobile first, cloud first to a world made from an intelligent cloud and an intelligent edge,” said Nadella.

As computing becomes more distributed, developers will not be able to write software bound to one virtual machine, according to Nadella. “Serverless computing will be the core of distributed computing, and this will change everything we do in Windows, Office 365 and Azure,” he added.

One of the companies using computing on the edge is Swedish industrial company Sandvik Coromant. It has built cloud applications to take all data from its machines to run predictive maintenance and time series analysis to identify the cause of anomalies. The company looked at how to shutdown its machines before they got damaged.

While this is being controlled by the Azure cloud, Microsoft has now developed an edge computing platform that would enable companies such as Sandvik to run machine monitoring applications on the machines rather than in the cloud.

Among the new products Microsoft unveiled at Build is Azure IoT Edge. This is a Windows and Linux cross-platform runtime environment, which Microsoft claims is able to run on devices smaller than a Raspberry Pi.

According to Microsoft, application logic can be tested in the cloud, and the same code can then be run on the edge device.


For casual readers, we really want this to happen.....intelligence on the edge for all devices, wearables, smartphones with edge intelligence would drive a need for eFPGA and
co processing......hardware acceleration. Some of these algos for AI are compute intense. QUIK has a old slide deck that has the compute intensity of various processes, voice
just starts to put the MCU into the yellow. 


Maybe one day they will add inference algos on to that slide? Don't really know where they fit into that yet.....

For casual readers

Inference on the edge has BIG dogs on record as wanting it. Facebook has it as a focus.

Now add in this from MSFT.

Open q to ponder...

Where will Inference on the edge run?