AWS (Amazon Web Services) released for general use its FPGA-based EC2 F1 instances in its cloud computing lineup in July, 2017. The EC2 F1 instance is based on Xilinx’s 16nm Virtex UltraScale FPGAs and people have been using this cloud-based hardware acceleration capability to speed up the execution of diverse tasks including the implementation of CNNs (convolutions neural networks), video transcoding, and genome sequencing. I’m certain there’s been some experimentation with high-frequency equity trading as well, but no one’s talking. Not to me, anyway.
Problem was, you could either get one FPGA (the so-called “f1.2xlarge” instance) or eight FPGAs (the “f1.16xlarge” instance. But like Goldilocks, some customers undoubtedly found the f1.2xlarge instance to be “too small” and the f1.16xlarge instance “too big.”
How do I know?
I know because AWS announced a “this one is just right” f1.4xlarge EC2 F1 instance today with two FPGAs.