What happens when the cloud meteorite hits the FPGA dinosaur earth?

What happens when the cloud meteorite hits the FPGA dinosaur earth?

Since 2017 and the announcement of the first FPGA instances publicly available in the cloud, many folks have been interested by this exciting trend.

While AWS  was the first major cloud service provider to offer FPGAs , many others companies have followed like Alibaba Cloud , Huawei CloudBaiduTencent, , Nimbix,  Azure and others.

I can witness that it is a subject that brings great excitement in the small world of FPGA industry.
But let’s take one minute to analyze the state-of-the-art before this event.

FPGA industry context:

I like this article published in 2015, still relevant to understand the FPGA and the IP historical context. To make a long story short, 3 actors are involved in this industry:

IP vendors:

An IP is a block of logic or data that is used to facilitate FPGA programming or acts as a library of hardware functions (for the software people like me)

Revenue model:
For any new deal signed, the IP vendor will charge an upfront premium cost called a license fee.
Based on factors such as the IP complexity, cost of development, target applications and expected volumes, a license fee can vary from 10k$ to 1000k$.

Board vendors (Intel FPGA & Xillinx)

They specify and build the physical chips and provide a unique range of tools to program them.

Revenue model:
From my analysis, the sale of the chips mainly generates revenue.

Solution vendors:

Solution vendors build complete system that are designed for end users. In order to achieve their goal, they need to provide the software and hardware components. It might require the development of their own IPs or the need to buy existing ones and the development of low-level (driver) and high-level (API) software layers.

Revenue model:
Revenue generated by the sale of the solution to the end users. It includes the software solution ( bitstream + software) or can be a package, a hardware blackbox providing a function (i.e compression, encryption etc …).
Here the end user is a large company for which you tuned the application, and who orders a large number of application packages. Still this end user is in charge of the maintenance and deployment of the solution.

 

Cloud context :

Now let’s talk about the cloud industry.

Let’s assume you’re an entertainment company providing movies and series to your customers. You’ve implemented a Cloud computing models that let you use virtual servers (instances) to transcode movies into various formats, depending on the end user characteristics.

Lucky you! Your company thrives but you’re reaching the limit of your Virtual Server (instances) renting models. This limit can be related to the cost increase due to maintainability, over capacity and scalability issues, but also rentability and competitiveness issues or technical evolution ( speed, latency ).

As a cloud user, your first impulse is to search on the marketplace for an appropriate solution that solves those issues with a higher level of satisfaction technically and economically.

If you are lucky, you will find something that fits your needs: let’s call it SaaS (Software-as-a-Service) video transcoding solution that better integrates to your system, speeds up your processing activities and overall performance while reducing the operating costs by managing the deployment, maintenance and scalability for you.

As an end user, your main responsibility is to pay for the usage of this service, keeping the control on process and without investing upfront costs.

Of course this is a caricature of the players but revels interesting challenges and highlight the business model gap between these 2 industries.

Industries gap FPGA vs Cloud

FPGA Industry Cloud Industry

End User

Large company

Medium to VeryLarge

Application expected revenue model

Fees + royalties

Pay-per-use

End user Skill

FPGA + Software

Cloud Software

Application developer skill

FPGA

FPGA + Software + Cloud

Product orientation

Hardware ecosystem

Software ecosystem

Product Usage

Using driver, dedicated hardware library

User interface or rest API

So why this eye-catching article title : “WHAT HAPPENS WHEN THE CLOUD METEORITE HITS THE FPGA DINOSAUR EARTH?” Mainly because many modifications will need to happen for the FPGA industry to move into the new world of the cloud industry. What will be the evolution? Which new FPGA species will emerge? Gladly some folks already started to provide appliances and facilitators for solution vendors who want to move to this new world. I have in mind :

Those solutions are promising and I look forward to discovering what they can hold for us and expose them to you in future posts.

Stay tuned …

 

 

Related articles & sources  :

By Jean-Michel Frouin on 10 April 2019