Portfolio optimization software now runs 23 times faster on graphics processors

A research team built a faster algorithm for solving large investment portfolio problems on graphics processors, cutting runtime from 65 seconds to 3 seconds on a real 5,440-asset dataset. This means financial firms can now rebalance massive portfolios in near-real-time instead of waiting minutes.

For decades, the math behind optimal investment portfolios has been computationally expensive — you need to juggle thousands of assets and their relationships simultaneously. This paper shows that modern graphics processors can do that work fast enough to be practical. The speed gain matters because portfolio rebalancing is a frequent operation; if it took minutes, you'd do it weekly or monthly. If it takes seconds, you can do it daily or in response to market moves. The limiting factor shifts from computation to something else entirely — the researchers note that after compression, the remaining bottleneck is no longer calculating matrix products, but solving the structured optimization problem itself.