March 18th, 2019 by Michael Rink
NetApp & NVIDIA Support AI Workloads
Throughout 2018, the momentum of machine learning advances steadily increased. With NVIDIA's GPU Technology Conference for 2019 coming up, this is a good time to review how NetApp and NVIDIA have contributed to advances in machine learning and AI over the past year and what is in store for this year.
Previously, NetApp’s ONTAP AI combined NVIDIA's DGX-1 supercomputer, NetApp's AFF A800 storage, and Cisco's network appliances to provide a single converged appliance solution with a thoroughly tested architecture. In the coming year, the DGX-1 will be swapped out for the NVIDIA DGX-2, which reportedly offers ten times the power of the first generation system. NetApp has announced that they will be sharing real-world AI centric benchmarks for the new appliance next week at the show; something to look forward to. The DGX-2 platform leverages the full NVIDIA GPU Cloud Deep Learning Software Stack to get the most out of its 16 fully interconnected GPUs. NVIDIA says it provides 2.4 TB/s of bisection bandwidth. The AFF A800 is the same component as was used in last year's appliance, and presumably, so are the Cisco components.
The new generation will also be added to NVIDIA's DGX-Ready Data Center program. The program gives customers access to data center services through a network of colocation partners. ScaleMatrix provides four data centers serving the U.S. as part of this program. Those four datacenters will include DDC liquid cooled cabinets that can support up to 52kW of power load in a single 45U cabinet. The cabinets are capable of holding the DGX-2 ONTAP AI appliances.
A number of artificial intelligence companies have either partnered with NetApp or use ONTAP AI. Allegro.ai uses NVIDIA DGX stations for their computer vision platform. Parabricks has been using ONTAP AI to shrink the time to perform secondary analysis of genomic data from a couple of days to under an hour. The H20 open source Driverless AI platform has been working with NetApp to make advanced data science and AI more accessible.