JP Morgan and NVIDIA have announced that NVIDIA’s (graphics) processors have been used successfully to accelerate derivative-focused risk computations by about 40X, when compared to running solely on classic processors (CPUs). Right now, JP Morgan computational center runs with a hybrid setup that includes both CPUs and GPUs.
Traditionally, those computations were run overnight so that useful data could be used for the next trading day. Now, the same results can be obtained in minutes, rather than hours, and in an rapidly changing environment, this is something that HP Morgan was willing to spend money on.The idea of using graphics processing hardware to perform non-graphics computations is not new. After all, modern GPUs have hundreds of computing units, and as long as the work can be split into independent tasks that don’t have too many conditions, it is possible to get hundreds of computing units working on a set of data. That is the real power of massively parallel computing.
For NVIDIA, this is great for two reasons: First, the financial sectors and other areas targeted by its Tesla products typically are computationally “unbounded”, which means that it’s never fast enough, which means that customers are willing to upgrade regularly. Secondly, enterprises are flush with cash, and looking for ways to invest on something that generate an immediate return on investment, and boosting the computing speed by 40X sounds just like it. [press release]