The AWS (Amazon Web Services) branch of Amazon (NASDAQ: AMZN) recently announced that it was going to be launching a brand new instance type called F1 which will come with FPGAs (Field-Programmable Gate Arrays). The promise here is that you will be able to reprogram the chips on the move and tune them with particular applications which would make them quicker than many regular CPU/GPU combinations.
The new instances were first made available in the form of a preview to the US East region, and as time goes by, they will become available to other regions. The pricing for these instances hasn’t been announced yet, unfortunately.
Even though they aren’t mainstream yet, FPGAs have become quite easy to program recently, which is why they are now finding their way into so many services. When they will be readily available in the cloud, it will mean more and more developers will begin experimenting with them.
4K and HD
The CEO of Amazon Web Services Andy Jassy had said that they have always tried their best to take anything that they can use themselves and share it with their customers.
The new instances are going to most probably be used for 4K video and HD processing as well as imaging and machine learning. For example, Microsoft went all out with FPGAs to create the backend of their AI services even though Google decided to take the more expensive route, and use specialized chips instead. Since these chips can easily be reprogrammed, it is very easy for you to switch context in applications too. This means you will be able to process raw images at one moment and then reconfigure the FPGA in seconds so that it can analyze a new image within milliseconds.
Among all of the companies that AWS has worked with in order to test the new instances, is this one company called NGCodec. This company used its RealityCodec for the AR/VR processing of the new instances within 4 weeks.
Ideally, this will help developers run complex video processing which is required for running AR or VR head-mounted displays through the cloud instead of doing so on a device. FPGAs have a major advantage over GPUs when it comes to codec use since it involves tons of decision making which the GPU prefers deferring to the CPU.
Amazon uses Xilinx chips for this. They are the last major FPGA manufacturer who is still independent. Programming these FPGAs is very hard and there is no sign that Amazon is planning on releasing tools that are going to help make it easier. There is going to be a development kit made available with it though, and a machine image which the developer will be able to use in order to begin with the new instances.
Xilinx on its own has also been releasing tools that make it much easier to use the more common languages such as C++ and C for programming the FPGAs. This is the same thing that the company had used for designing the decoder for all of the F1 instances.