Azure Essentials: High Performance Computing (HPC) options
Articles,  Blog

Azure Essentials: High Performance Computing (HPC) options


(swooshing) – Welcome to Azure Essentials. In the next few minutes, we’ll explore the services in Azure
to run high performance computing applications and solutions. Azure can deliver all
sizes of computation. From enabling powerful visual simulations and engineering scenarios,
to dynamic rendering for GPU intensive work, to scientific and data scenarios using
analytics languages like R. When it comes to high performance compute, we know that some jobs may
require high CPU performance, while others may need a
high memory to core ratio, dedicated GPUs or even
dedicated super computers. You can mix and match your infrastructure, and change it as your
business needs change. For larger scale solutions,
Azure also offers low latency InfiniBand interconnects for running massive HPC
jobs across multiple VMs. And you can use dedicated infrastructure with Cray supercomputers for
extremely demanding workloads. So, let’s walk you through
your options in more detail, and along the way, we’ll point out a few common scenarios that can take advantage of high performance computing in Azure. Starting with your infrastructure options, Azure now offers virtual chains with fast CPUs for
simulations or analytics, large memory for passing databases within memory capabilities,
as well as GPU support. Of course all options
support both Windows, and Linux-based workloads, as well as a variety of open-source development, automation, management platforms. GPUs can of course be used
for improved visualization, but the speed of the GPU also
enables faster computation. Now with graphics-intensive scenarios, Azure’s N-series virtual
machines are optimized specifically for native-like
access to GPU resources, offering near bare-metal
performance levels. And V-series VMs are great for running computer-aided design, resource-intensive visual simulations, allowing engineers to access powerful Azure-hosted compute from
just about any work station, and without needing to move terabytes of modeling data between
end points and servers. Microsoft has partnered with Nvidia, to offer their Tesla V100
PCI Express-based GPUs, and V100 SXM GPUs, so scenarios like game development or video production provide dynamic and photo
realistic rendering, as you can see here in this simulator. You can also see the high frame rates, and graphics performance. Beyond graphical scenarios, Azure provides ND-series virtual machines for artificial intelligence workloads, like model-training and inference. For high performance computing workloads like computational chemistry
and fluid dynamics simulations, you can use NC-series, and H-series VMs. Now because Azure’s the only public cloud to provide InfiniBand and RDMA to connect thousands of instances across multiple GPUs and compute nodes, these services provide extremely fast compute times, at low latency. Especially compared to
CPU only architectures. And if your jobs require the massive scale of a Cray supercomputer, we’ve got an exclusive
partnership with Cray that allows you to get a
dedicated supercomputer running within your Azure virtual network. Now you’ve seen how Azure
provides infrastructure for all of your big compute workloads. Now let’s look at how Azure makes using those resources easier. Firstly, Azure Batch can
run complex batch processes, which can reason over large amounts of data and compute
resources, as a service, without the need to manage individual VMs, so you can just focus
on running the workload, versus provisioning
the computer to run it. Within Azure Batch, we also have accelerators for AI, and rendering. Now to provision, monitor, auto-scale, and manage the lifecycle
of HPC compute clusters, we also have Azure Cycle Cloud. This allows IT groups to provide
managed clusters in Azure. Of course, you can use VM scale sets, or Azure resource manager
templates to help you deploy, manage, and scale your solutions as well. Once you’re up and running, Azure offers flexible consumption, you can
change your infrastructure as your workload or business needs change. Here’s an example using Cycle Cloud, to quickly change the virtual
machine type you’re using. Azure also provides granular monitoring of usage and associated costs. In addition, low priority
VMs are allocated from our surplus compute capacity,
this means you can run batch workloads requiring a large amount of compute power at up to an 80% discount. So that was a quick
run-through of your options for high performance computing in Azure. Azure has high performance
computing options to fit your specific
needs across all scales. If you’re interested in
the comprehensive set of compute options in Azure,
there are dedicated topics for IaaS and modern compute
in Azure Essentials. Of course, we’re constantly
adding new topics on Essentials, so please keep checking back for more. You can also learn more at
our hands-on learning series, at the link shown, thanks for watching. (metallic swooshing)

2 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *