Recently developed an interest in the topic of programmable matrix FPGA, which AMD can hook to their processors for the needs of one of the major customers.
Jensen Huang began with the recognition of the fact that the FPGA used at NVIDIA for a long time to emulate the developed products. And there are applications where these matrices are effective, but gives a very narrow specialization. Graphic processor is capable of operating at higher frequencies, and perform 100 times faster. Thus the creation of matrix leaves the work of several engineers and to develop GPU employs thousands of engineers. It is the proportion of labor costs, according to the head of the company and determines the result - graphics processors not only faster, but also more universal, if we talk about speeding up calculations.
Use FPGA's data centers, according to Juan Jensen, is not very reasonable. The graphics processor is able to cope with the diverse and demanding workloads better than the programmable matrix. Therefore, NVIDIA did not see a better alternative to computing accelerators based on GPU.