Investment banks have been spending millions of dollars to find the best way to assess risk and to buy and sell the right stocks at just the right time. Now that Nvidia?s GPU accelerated technology is available to investment banks there has been an arms race between investment banks to find the best IT infrastructure to facilitate faster and more accurate trading.
In the past, most of these risk calculations computers were powered mostly by x86 machines from either AMD or Intel. And to be quite honest, both companies still have a relatively important role; for now. Nvidia today announced that their partnership with JPMorgan in helping them implement Nvidia TESLA GPUs in their risk calculation computers. They stated that in real world usage, they saw an increase of 40 times in the speed of their risk calculations while simultaneously reducing the total cost of ownership. JPMorgan?s equity derivatives group did this by adding Nvidia?s TESLA M2070 GPUs to its datacenters.
Nvidia stated that more than half of their risk calculations had moved from CPU based systems over to hybrid CPU/GPU solutions. This is still a nod to the fact that Nvidia simply cannot sell their GPUs without Intel or AMD selling at least one processor. At the current moment, most servers will run between 1 and 4 GPUs per CPU and as such many companies see a significant boost in performance and reduction in operating costs. This is why Intel is moving towards implementing Larrabee 2.0 aka Knight?s Ferry MIC. At this moment, Intel has no actual products to show so Nvidia is effectively running the GPGPU/cGPU show. These TESLA GPUs that have been implemented inside of JPMorgan have been deployed in multiple data centers across JPMorgan?s Compute Backbone infrastructure. Because of this unified IT infrastructure and the ability for nearly any office to utilize these powerful datacenters, the systems have resulted in utilization rates of 70 percent on the GPUs while also running 24 hours a day.
JPMorgan reported to Nvidia that they were implementing this setup as part of a three year plan to reduce their costs of risk calculation by 75 percent. This also doesn?t take into account that swapping out GPUs for upgrades is much easier than swapping out CPUs as many newer CPUs are not socket compatible with older CPUs. This is a thing to take into consideration because if companies can drop-in upgrade GPUs (PCIe) then they don?t have to spend money on a new CPU or motherboard if the GPUs can bear most of the load and show a significant improvement in speed. This could also hurt companies like Intel and AMD who traditionally expect there to be an enterprise upgrade cycle of server chips every few years.
The most important factor, though, of using these GPUs is that it simply reduces their wait time for these calculations down to minutes rather than hours. This reduction not only improves the workflow of the employees, but it also gives them a competitive advantage because they are basically able to make decisions on the fly rather than having to wait to make those decisions. Some investment opportunities may need to be decided on within minutes, not hours. As a result of JPMorgan?s efforts in GPU utilization in Risk Assessment Calculations, they have actually won an award for their innovation in banking technology.