Microsoft wants to use AI to unlock the potential of HPC
By working with its partners, Microsoft has created the Azure HPC & AI Collaboration Centers program which funds collaborations with well-known institutions in an effort to develop and share best practices with both the HPC and AI communities.
The software giant is delivering this program in partnership with AMD, Intel and Nvidia and so far five Azure Collaboration Centers are in development with Duke University, Purdue University, Red Oak Consulting, Numerical Algorithms Group (NAG) and RedLine Performance.
Microsoft Azure is able deliver differentiated cloud capabilities that are purpose-built for the full spectrum of HPC & AI workloads including the most demanding use cases. However, at the core of the infrastructure are HPC & AI clusters which use HPC-class InfiniBand interconnect with genuine low latency (~1us) and high bandwidth typically around 200Gbps using the latest processor technologies from its partners in the program.
As a result of these performance advantages, Azure is able to show scaling of supercomputer workloads to 12 times more CPU cores than other clouds and to deliver cost-to-solution leadership over other solutions. At the same time, Microsoft has built HPC & AI expertise across the spectrum from engineering teams, to customer engagements to through-life customer support.
HPC and AI Collaboration Centers
The end goal of Microsoft and its partners’ new program is to demonstrate and share best practices surrounding how Azure HPC & AI capabilities can be used to unlock customer innovations and productivity.
The Duke University Azure HPC Collaboration Center, in partnership with Nvidia, plans to use Azure GPU capabilities to develop personalized hemodynamics simulations capabilities in the cloud with the aim of improving our understanding and treatment of coronary bifurcation lesions.
The Purdue University Azure HPC & AI Collaboration Center, in partnership with AMD, will demonstrate best practices for using the Azure HPC cloud to augment on-premises HPC systems such as Purdue’s AMD-based Anvil supercomputer.
The Read Oak Consulting Azure HPC & AI Collaboration Center, in partnership with Intel, will develop and showcase best practices for building HPC clusters on Azure using the chipmaker’s technology.
The NAG Azure HPC & AI Collaboration Center, in partnership with Nvidia, will be focused on developing best practices for deploying scalable machine learning.
Finally, the RedLine Performance Azure HPC & AI Collaboration Center, in partnership with Intel, will implement and demonstrate the National Weather Service (NWS) Finite Volume Cubed Sphere Global Forecast System (FV3GFS) workflow on Azure, from the Global Data Assimilation System (GDAS) through post-processing.
Microsoft plans to announce further collaboration centers over the coming months and we’ll likely hear about the advancements being made at the centers that have already been established soon.