JPMorgan server joins GPU 40 times times faster

Sina Science and technology news Beijing time August 5 afternoon, JPMorgan and Nvidia Thursday together demonstrated the benefits of increasing the GPU in the server.　　The investment bank is one of the industries most valued at computing speed, and while these companies have been secretive about how to improve server performance, JPMorgan has disclosed some information this time. The GPU, also known as a graphics processor, was used primarily to render ornate graphics in the game.　　While the x86 CPUs produced by Intel and AMD are still dominant, the GPU is better at special tasks. The NVIDIA GPU has been integrated into JPMorgan's servers to serve as an accelerator for x86 processors. The two companies say that in this way, peak performance can be as high as 100 times times that of using only CPUs when performing certain tasks.　　In a more typical case, JPMorgan has increased its speed by 40 times times, reducing its cost by 80% as the number of servers required has decreased. But such a measure would also create an additional cost: to fully play the role of the GPU, the software needs to be adjusted, which would discourage some computer users.　　But for JPMorgan's speed-chasing users, it's worth a try. JPMorgan's equity derivatives group will use the GPU to perform the so-called "risk Computing" (disorientated computation) mission. The sector has a large number of complex financial assets, each of which may be linked to a variety of underlying securities, and a variety of possible scenarios need to be evaluated to understand the potential changes in the value of these assets. For example, the economic downturn in some countries or the fall in stock markets will have consequences.　　More powerful computing power allows the department to add more variables to each of these possibilities, bringing them closer to the actual situation, while also calculating more possibilities within a certain amount of time. Other Wall Street banks are also assessing the effects of the GPU, but not all will use Nvidia's products.　　AMD is also trying to drive similar functions on the GPU. At the same time, JPMorgan and other investment banks are deploying a chip called FPGA (Field Programmable gate array, field programmable gate arrays) to make up for x86 chip deficiencies in part of the task. Intel usually does not focus on GPU development due to the complexity of programming. But the company is also designing a hybrid chip mic with multiple processors and plans to use it as an accelerator for mainstream microprocessors. (Ding Macro)