NVIDIA Kepler GPU Architecture, 2012
Jensen Huang presents NVIDIA’s Kepler compute architecture at GTC 2012. He compares the new SMX streaming multiprocessor with Fermi, focusing on a threefold gain in energy-efficient performance. Huang then explains Hyper-Q’s 32 concurrent work queues and dynamic parallelism, which lets GPU threads create further work for data-dependent, nested algorithms such as adaptive turbulent-flow simulation.
Passages
1:17 “We know that the ultimate performance of computing in the future is limited by energy. Whether you have a data center, the size of this convention hall, or a laptop, or a phone or a tablet, it doesn’t matter what level of computing we’re talking about.” Share
Search what Huang said from Claude, ChatGPT or your own code. See plans