NVIDIA is making some noise in the supercomputing world today by unveiling its latest Tesla K-Series, in which K stands for “Kepler”, which is the name of the micro-architecture used to build it. That is the same architecture used in NVIDIA’s most recent graphics processors (GPUs) which is now making its way to the supercomputing world.
Unlike traditional GPUs, the Tesla Series is built to take advantage of the massive parallelism of graphics processors to accelerate general computation tasks. In fact, the Tesla cards don’t even have a video-out port which could remind us of their graphics cards cousins.
With 18688 NVIDIA Tesla K20X GPUs, the Titan supercomputer from the Oak Ridge National Laboratory has grabbed the number one spot in the global Top 500 list. The Linpack benchmark clocked in at 17.59 petaflops.
It was only a matter of time before it happens as GPUs can increase their overall performance not only by scaling the clock, but more importantly by scaling the number of “cores” or atomic computing units inside the chip. GPUs started modestly with one computing core associated with one texture unit, but in 15 years, they now feature hundreds of cores in a single chip.
NVIDIA’s Kepler architecture has been optimized to pack even core per square millimeter, which ultimately translate into more processing power per chip. Now the question is: how long will it take for GPUs to completely overrun the Top 500 list? Let’s wait and see…