Building Faster Cryptography with Carryless Multiplication in NVIDIA CUDA 13.3 | NVIDIA Technical Blog

What this post added

Introduced the `clmad` PTX instruction in CUDA 13.3, providing hardware acceleration for carryless multiplication on NVIDIA Ampere and newer GPUs. Demonstrated significant performance improvements for GHASH (up to 18.8x on B200) and sum-check protocols for zero-knowledge proofs (3-13x speedup) by leveraging this new instruction, enabling faster cryptographic and coding-theoretic workloads.