Fig. 3
From: Parasail: SIMD C library for global, semi-global, and local pairwise sequence alignments
Performance dependency on number of threads and query length for global alignments. The speed in billion cell updates per second (GCUPS) of striped SSE4.1 vectors (red), prefix scan SSE4.1 vectors (orange), striped AVX2 vectors (black), and prefix scan AVX2 vectors (green), as well as libssa using SSE4.1 (purple) and AVX2 (gray) instruction sets, using a variable number of threads and queries of varying length. a Number of threads ranging from 1 to 24 and the 375 residue long P07327 query sequence. b Query sequences ranging from 24 to 5478 residues in length and 24 threads. c Query sequences of varying length and 1 thread. All scales are logarithmic. The BLOSUM62 matrix and gap open and extension penalties of 11 and 1, respectively, were used in all cases