This blog post was authored by the Azure High Performance Computing Team.
Every year at Supercomputing, over 12,000 researchers and practitioners come together to push the state of the art in high-performance computing forward. This year is no exception, and we’re excited to join the global HPC community in Dallas, Texas next week.
In booth #3442, Microsoft will showcase the work we have done to address key cloud migration blockers including price performance, simple/performant file access for both MPI and massive-read workloads, and challenges for hybrid HPC enterprises managing across cloud and on-premises clusters.
Over the last year, we have worked to deliver the most complete set of integrated services and solutions to enable customers to get started faster, run HPC workloads cost-effectively, easily migrate or burst workloads to Azure with little risk, and start to control costs/access across the hybrid HPC enterprise. These capabilities are highly scalable, deeply integrated, and very efficient in price-performance and time-to-value.
Specifically, at our booth we’ll show off these Azure HPC solutions:
- Price-performant cloud infrastructure Virtual Machines: New releases of HPC Inifiniband interconnected VMs for leading price/performance and MPI scale:
- HB VMs with AMD EPYC processors, 60 cores, and 4GB RAM per core (Learn more)
- HC VMs with Intel Xeon Platinum processors, 44 cores, and 8GB RAM per core (Learn more | Sign up)
- Price Performant bare-metal HPC infrastructure: Cray CS Linux clusters, Cray XC supercomputers that are built to customer spec, and match on-premises cost for high-utilization workloads
- High-performance HPC storage: Cray ClusterStor file systems capable of feeding supercomputer-scale compute workloads, on the Azure Network
- Simplified HPC Cluster hybrid management: Azure CycleCloud for easy at-scale orchestration, access management, error-handling, and cost control with templates for traditional HPC clusters, parallel file systems, and common industry workloads
- Hybrid, performant file access: Avere on Azure leading low-latency, high-performance file caching for hybrid & cloud-only workloads
- Advanced Machine Learning capabilities: New NDv2 GPU VMs for the latest in scale-up AI training using NVidia NVLink technologies with a dense 8-way Volta GPU configuration (Learn more | Sign up)
- Simplified Remote Visualization: New NVv2 GPU VMs with GRID licensing included (Learn more | Sign up)
- Cloud Native HPC Workloads at Scale: Azure Batch cloud native job scheduling for HPC applications where Batch service manages the underlying cloud-scale complexity and you focus on your application needs.
- Preview the New “Sentinel” Cray in Azure: Cray and Microsoft are proud to announce the availability of Sentinel, the Cray in Azure invite-only preview system. If your organization needs considerable HPC capability, apply for a pilot on Sentinel to experience all of the above capabilities in Azure.
Microsoft also has considerable technical content this year, with presentations including:
- A Quantum Future of Computation invited talk session with Matthias Troyer on Thursday, November 15, 2018 from 8:30 AM – 9:15 AM Central Time.
- Panel Discussion: HPC in the Cloud Exhibitor Forum Session at the Intel Booth (D171) on Tuesday, November 13, 2018 from 1:30 PM – 3:00 PM Central Time:
- Track 1: 1:30 PM – 2:00 PM Central Time | Extending On-Premise HPC to the Cloud | Presenter: Gabriel Broner
- Track 2: 2:00 PM – 2:30 PM Central Time| Enabling HPC and Deep Learning Workloads at Extreme Scale in the Cloud | Presenter: Bill Bryce
- Track 3: 2:30 PM – 3:00 PM Central Time | What Would You Do with a Million Cores of Compute Capacity? | Presenter: Ian Colle
- Applications of Deep Learning in Industry and Research Birds of a Feather Session on Wednesday, November 14, 2018 at 5:15 PM – 6:45 PM Central Time.
- Session Leader: Lucas Wilson
- Additional Session Leaders: Antonio Cisternino, Matthew Grover, and Alex Sutton
- Benchmarking Scientific Reconfigurable/FPGA Computing Birds of a Feather Session on Wednesday, November 14, 2018 at 12:15 PM – 1:15 PM Central Time.
- High Level Programming Languages for Quantum Computation Birds of a Feather Session on Tuesday, November 13, 2018 at 12:15 PM – 1:15 PM Central Time.
- Session Leader: Torsten Hoefler
- Additional Session Leader: Matthias Troyer
- “Reconfigurable Computing for HPC: Will It Make It This Time?” Panel Discussion on Wednesday, November 14, 2018 at 1:30 PM – 3:00 PM Central Time.
- Moderator: Franck Cappello
- Panelists: Torsten Hoefler, Kentaro Sano, Maya Gokhale, Andrew Putnam, Kazutomo Yoshii, and Ronan Keryell
We’re also proud to be a sponsor of the Student Cluster Competition, which will include a cloud GPU access, along with tables at the Student Career Fair. We look forward to seeing you in Dallas this week!