According to our estimations, with MatLogica AADC, you will be able to speed up some of your models to up to {FasterFactor}X! This means your cloud costs can go down by up to {costSaving}K$ per year. Your code can be simplified, meaning {simplificationFactor}% faster time to market for new models.
When modeling these savings, we've considered your answers to the questions, the general functionality of MatLogica AADC, and our past experience in working with organisations similar to yours. These factors include:
- The way of calculating sensitivities:
- AADC code generation approach enables producing all sensitivities faster than the original program computes the PV itself. Alternative AAD tools will slow down the program by a factor of 5-10.
- Manual differentiation is hard to maintain, especially when it comes to higher-order greeks; Bump-and-revalue is slow!
- The use of multithreading: typically, financial institutions are reluctant to use multithreading; AADC enables safe and scalable execution on multi-core systems;
- Vectorisation: AADC enables automatic vectorisation; AVX-2 delivers processing 4 samples of the data in one cycle and AVX-512 is another ~1.7X faster!
- JIT optimisation: AADC kernels are generated at JIT when a specific task configuration is known, which allows us to perform additional optimisations to deliver this performance boost!
Please note that each case is unique and this is a rough assessment comparing your answers to similar organisations.