Inside the Microsoft FPGA-based configurable cloud

Play Inside the Microsoft FPGA-based configurable cloud
Sign in to queue

Description

Microsoft has been deploying FPGAs in every Azure server over the last several years, creating a cloud that can be reconfigured to optimize a diverse set of applications and functions. This configurable cloud provides more efficient execution than CPUs for many scenarios without the inflexibility of fixed-function ASICs at scale. Today, Microsoft is already using FPGAs for Bing search ranking, deep neural network (DNN) evaluation, and software defined networking (SDN) acceleration. Azure’s FPGA-based accelerated networking reduces inter-virtual machine latency by up to 10x while freeing CPUs for other tasks. We describe Microsoft’s cloud FPGA architecture, show how these applications are using it, show live demos of the performance that FPGAs provide, and discuss possible uses. 

Tags:

Azure, Bing

Day:

2

Session Type:

Breakout

Code:

B8063

Room:

WSCC Hall 6E (1215)

Embed

Download

The Discussion

  • User profile image
    CarlD

    Awesome talk, Mark.

  • User profile image
    Mazen Ezzeddine

    Thanks for the talk, Interesting.

    However, very high level description of the design, deployment and infrastructure. For instance., an end to end client job submission (use case) is not discussed. Say, a client want to train a custom DNN how many FPGAs are granted, how data is partitioned and stored near FPGA for subsequent training iterations, where client data is stored in CPU memory, in FPGA memory what agent keep track of data locations, how accelerated microservices call one other what is the API/protocol is the cpu involved if not how, a single client query could actually be a DAG of microservices how this DAG is mapped to FPGAA microservices. Different cloud clients e.g. big data analytics client, video streaming client, scientific computing client etc..

    Thanks.

Add Your 2 Cents