Table 2.
Optimization | Speed-up compared to minimac |
|
---|---|---|
minimac2 | minimac2 (four cores) | |
Data locality | 4.5× | 13.2× |
Vectorization | 3.8× | 12.3× |
Adaptive precision | 1.5× | 5.8× |
Overall | 18× | 55× |
All experiments were run on a server with four 2.4 GHz Intel Xeon, 128 GB of RAM, gcc 4.7.2, and OpenBLAS 0.2.11. minimac(2) required a maximum of 1.1 GB (2.8 GB using four cores) memory to impute the genome in 5-Mb chunks (including 0.5 Mb overlaps, total 6 Mb, up to 110 350 variants).