"Binius’s FPGA prototypes have shown that binary field multiplications can be performed in one clock cycle, making them exponentially more efficient than traditional prime field operations like Mersenne-31, which require 49 clock cycles"

"Binius’s FPGA prototypes have shown that binary field multiplications can be performed in one clock cycle, making them exponentially more efficient than traditional prime field operations like Mersenne-31, which require 49 clock cycles"

https://0xmilica.com/post/binius/

That’s sick! It’s fine if it’s not consumer hardware - I’d expect that outsourcing large proofs to an untrusted server will see some breakthroughs in upcoming years. For zkVMs this is huge, for consumer hardware one can rely on optimized circuits for specific tasks.

Building a verifiable future. Previously research at @ucberkeley + @a16zcrypto, PhD @TUM

https://ernstberger.xyz

In the blockchain context, it's totally reasonable to ask someone who is trying to craft a private transaction to do a proof with private inputs themselves on consumer hardware even if it takes a few minutes, while most users just offload their transactions to some proving service.

I see a world where large proving farms are mostly just aggregating state transitions without needing to see any of the higher level context (merkle root was this, and is now that)

Users (and shared services) hold their own private data that allows them to craft merkle state attestations as needed. Provers only see a list of updated state (access lists), the chain only sees a new merkle root.