CloudCrack is an NVIDIA GPU-accelerated cryptanalysis suite written in CUDA, NVIDIA's massively parallel concurrent programming language. CloudCrack contains custom CUDA multiprecision math libraries for storing a large target RSA modulus n in shared GPU memory, with each GPU core working as a parallel factoring process to break the target modulus.
There are currently two versions of CloudCrack: REVA, which implements the Greatest Common Denominator (gcd) function on the GPU itself; currently there is a bug in the Montgomery math routines in the REVA gcd however. REVB includes reduction sieve performance enhancements but with the gcd function implemented on the host CPU, which requires about 25% of the PCIe bus bandwidth to shuffle targets from the GPU to the host CPU for the gcd test. Although functional, this split GPU/CPU configuration is suboptimal for highly dense GPU clustering hardware that utilizes a shared PCIe bus across multiple GPU cards such as the Dell C410x PCIe Expansion Chassis. Ideally REVC will include all of the performance enhancements inherent to the REVB fork, with a GPU-localized gcd like the architecture of REVA..
Read more: CloudCrack