RAxML-VI-HPC is a fast and efficient program for maximum likelihood (ML) phylogenetic analysis, capable of handling large datasets with thousands of taxa. It is significantly faster than previous versions of RAxML, with performance improvements ranging from 2.7 to 52 times. It uses technical optimizations, a modified search algorithm, and the GTR+CAT model as a replacement for GTR+Γ, which is more computationally intensive. RAxML-VI-HPC is parallelized with MPI to enable parallel processing on PC clusters. It can compute ML trees on large alignments with up to 25057 taxa. It also supports mixed models for multi-gene alignments and constraint trees. RAxML-VI-HPC outperforms other ML programs like GARLI, IQPNNI, MrBayes, and PHYML in terms of inference speed, memory usage, and log-likelihood values, especially for large datasets. It is particularly efficient for datasets with more than 4000 taxa. The GTR+CAT model is more efficient than GTR+Γ, making it suitable for large alignments. GARLI, despite being a genetic algorithm, performs well on smaller datasets but is slower than RAxML-VI-HPC. MrBayes also performs well, but its Bayesian approach is less suitable for comparison with ML-based methods. IQPNNI and PHYML suffer from inefficient implementations, leading to high memory usage and slower performance. PHYML uses NNI moves that only explore a small part of the search space. RAxML-VI-HPC's technical optimizations and efficient implementation contribute to its superior performance. The study highlights the importance of technical implementation in phylogenetic analysis and the potential for significant performance improvements through algorithmic advancements. The new version of RAxML incorporates efficient technical optimizations, parallel implementations, and mixed models. It can find better trees with lower memory consumption than the best competing program. Future work will focus on developing new methods for rapid bootstrapping, as full biological analysis requires a large number of bootstraps. The study underscores the importance of technical implementation in phylogenetic analysis and the potential for significant performance improvements through algorithmic advancements.RAxML-VI-HPC is a fast and efficient program for maximum likelihood (ML) phylogenetic analysis, capable of handling large datasets with thousands of taxa. It is significantly faster than previous versions of RAxML, with performance improvements ranging from 2.7 to 52 times. It uses technical optimizations, a modified search algorithm, and the GTR+CAT model as a replacement for GTR+Γ, which is more computationally intensive. RAxML-VI-HPC is parallelized with MPI to enable parallel processing on PC clusters. It can compute ML trees on large alignments with up to 25057 taxa. It also supports mixed models for multi-gene alignments and constraint trees. RAxML-VI-HPC outperforms other ML programs like GARLI, IQPNNI, MrBayes, and PHYML in terms of inference speed, memory usage, and log-likelihood values, especially for large datasets. It is particularly efficient for datasets with more than 4000 taxa. The GTR+CAT model is more efficient than GTR+Γ, making it suitable for large alignments. GARLI, despite being a genetic algorithm, performs well on smaller datasets but is slower than RAxML-VI-HPC. MrBayes also performs well, but its Bayesian approach is less suitable for comparison with ML-based methods. IQPNNI and PHYML suffer from inefficient implementations, leading to high memory usage and slower performance. PHYML uses NNI moves that only explore a small part of the search space. RAxML-VI-HPC's technical optimizations and efficient implementation contribute to its superior performance. The study highlights the importance of technical implementation in phylogenetic analysis and the potential for significant performance improvements through algorithmic advancements. The new version of RAxML incorporates efficient technical optimizations, parallel implementations, and mixed models. It can find better trees with lower memory consumption than the best competing program. Future work will focus on developing new methods for rapid bootstrapping, as full biological analysis requires a large number of bootstraps. The study underscores the importance of technical implementation in phylogenetic analysis and the potential for significant performance improvements through algorithmic advancements.