Modular Mojo claims to be over 36,000 times faster than Python for AI workloads
Modular Mojo is a new programming language designed for AI developers that is said to combine the usability of Python with the performance of C with over 36,000 times the performance of Python on a matrix multiplication workload. Modular Mojo programming language was not in the initial plan of the company but came about when the company’s founders – who focused on building a platform to unify the world’s ML/AI infrastructure – realized that programming across the entire stack was too complicated and also ended up writing a lot of MLIR (Multi-Level Intermediate Representation) by hand. The “over 36,000 times speedup” claim comes with the matmul.py script performing a 128×128 matrix multiplication in Python with a throughput of 0.00215 GFLOP/s and another script doing 512×512 vectorized + parallelized matrix multiplication in Mojo at 79.636 GFLOP/s. The claim looks dubious and that’s odd they used different matrix sizes, but some are […]