Sprang precis på artikeln Fast Implementations of AES on Various Platforms (pdf) av Joppe W. Bos, Dag Arne Osvik och Deian Stefan som beskriver flera nya mycket snabba implementationer av blockkryptot AES-128. Artikelns sammanfattning säger det mesta:

This paper presents new software speed records for encryption and decryption using the block cipher AES-128 for different architectures. Target platforms are 8-bit AVR microcontrollers, NVIDIA graphics processing units (GPUs) and the Cell broadband engine.

The new AVR implementation requires 124.6 and 181.3 cycles per byte for encryption and decryption with a code size of less than two kilobyte. Compared to the previous AVR records for encryption our code is 38 percent smaller and 1.24 times faster.

The byte-sliced implementation for the synergistic processing elements of the Cell architecture achieves speed of 11.7 and 14.4 cycles per byte for encryption and decryption.

Similarly, our fastest GPU implementation, running on the GTX 295 and handling many input streams in parallel, delivers throughputs of 0.17 and 0.19 cycles per byte for encryption and decryption respectively. Furthermore, this is the first AES implementation for the GPU which implements both encryption and decryption.

Artikeln ger bra information om optimeringar som gjorts för de olika arkitekturerna samt jämför med andra implementationer. Jag gillar även det faktum att man faktiskt nådde 59 Gbit/s(!) på en NVIDIA GTX 295, 1.24GHz.