0
votes

Coming from AWS, I am completely new to Azure in general and to Cloud Services spesifically.

I want to write a python application that leverages GPU on Azure in a PaaS architecture (Platform as a Service). The application will hopefully be deployed somewhere central, and then a number of GPU enabled nodes will spin up and run the application until it is done before closing down again.

I want to know, what is the shortest way to accomplish this in Azure?

Is my assumption correct that I will need to use what is called Cloud Services with a worker role, or will I have to create my own infrastructure based on single VMs running in IaaS?

1
All SKUs are documented in detail on the Azure pricing page. gpu is only available on Virtual Machines (and yes, worker role instances are, under the hood, stateless Windows virtual machines, but there is no GPU-based sku for these, as it's a very different offering than Virtual Machines). But as far as the rest of your question - not sure what you mean, about the shortest way to accomplish something.David Makogon
VM Scale Sets can run on the N-series. Scale Sets are IaaS with autoscale (which you can hook to different triggers like CPU, memory, queues, etc) - docs.microsoft.com/en-us/azure/virtual-machine-scale-sets/…evilSnobu

1 Answers

1
votes

It sounds like you created an application which need to do some general-purpose computing on GPU via Cuda or OpenCL. If so, you need to install GPGPU driver on Azure to support your Python application, so the Azure NC & NV Series VMs are suitable for this scenario like on AWS, as the figure below from here.

enter image description here

Hope it helps. Any concern, please feel free to let me know.