High-Performance Engineering for SNARKs, w/ Jim Posen

In this module, Nicolas Mohnblatt and Jim Posen, CTO at Irreducible, discuss high performance engineering for SNARKs. They cover identifying and overcoming computational and memory bottlenecks, leveraging parallelism through software and hardware engineering techniques, and employing SIMD instructions for operations like constraint evaluation, PCS opening, and hashing. Jim explains how modern CPUs and GPUs can be exploited for parallelism, introduces the concept of operation pipelining, and explores the potential of using custom hardware like FPGAs and ASICs. He also delves into the concept of recursion in SNARKs to handle large computations by breaking them into smaller segments and aggregating the resulting proofs. The discussion emphasizes the importance of efficient witness generation and the future of SNARK computations with custom hardware specific to cryptographic needs.