At the recent HPC Users Forum in Tuscon, AZ, NVidia was on-hand and gave a presentation on their internal HPC focuses. With lots of the usual buzz about CUDA downloads and specific codes ported with amazing performance boosts, they did offer a few new nuggets of information about their next system with Stacked DRAM and a future project they are targeting for 2018 called “Echelon”. Echelon is more than a chip, it’s an entire cabinet design that seems to aim for 4 Petaflops and 128Terabytes of ram per cabinet, with scalability up to the exascale.
Watch the video or view the slides at the link below.