====================================================================
Published: 1 December 2022
Tags: aws, fpga, optimization, rust
Tony Wu discusses a pain point that occurs "when transferring large amounts of data from a software application running on a typical x86 host machine to an FPGA for accelerated computation".
Some highlights:
- "Multi-Scalar-Multiplication is one of the major computational bottlenecks at the core of many Zero-Knowledge-Proof systems"
- "Xilinx/AWS provides a Linux-based driver running in kernel space that manages the low-level configuration of the DMA hardware on the FPGA"
- "Using CPU cores to transfer data usually has low latency, however it is very inefficient for large amounts of data and can result in low performance, so we would much rather use a DMA for the task"