Microsoft Retires HPC Pack: Windows Server HPC Migration Deadline Is August 27, 2027

The announcement appeared on Microsoft's Azure High Performance Computing blog on the Community Hub and was mirrored by other Microsoft-watching sites. Microsoft describes the product as a Windows‑based high‑performance computing (HPC) scheduler that enables customers to deploy, manage, and operate HPC workloads in on‑premises and hybrid environments. Argon Systems' copy of the post says HPC Pack is now entering a one-year retirement period that ends at End of Support on August 27, 2027.

The key dates, and a discrepancy to watch​

Microsoft's Community Hub announcement gives August 27, 2027 as the end-of-support and retirement date. The current Microsoft Learn page "HPC Pack 2019 retirement" matches that date. It was last updated on September 25, 2026.

Some Microsoft documentation shows a different date, though. The search-indexed version of that same Learn migration page said Microsoft will retire HPC Pack 2019 on August 30, 2027. Microsoft's Azure Virtual Machines documentation for HPC Pack 2019 also states that it is scheduled for retirement on August 30, 2027, and will no longer be supported after that date.

The most recently updated migration page says August 27, so the August 30 date looks like an earlier draft that Microsoft has since changed. Until every page agrees, plan around August 27, 2027. The last thing you want is to find out the deadline was three days earlier than the date on your project plan.

Section summary: Use August 27, 2027 as your deadline. Some Learn pages still list August 30, so don't treat the documentation as settled yet.

What "retired" means here​

Your clusters won't stop working on the deadline. According to the announcement, existing HPC Pack deployments will not be forcibly disabled; however, they will be considered unsupported. The Learn FAQ says the same: existing clusters will continue to run, but they will no longer be supported once the cutoff passes.

After that date, the Learn guidance says no further security or product updates will be released. Microsoft technical support will no longer be available. Documentation updates will be limited or discontinued. Microsoft lists three risks of running unsupported clusters:

• More security and compliance exposure
• Compatibility problems with newer operating systems or integrations
• Less ability to troubleshoot or fix problems

There is also no upgrade to wait for. The FAQ says plainly that there is no newer HPC Pack version. You should transition to supported Azure HPC solutions.

Support between now and the deadline​

Microsoft calls the support it offers until August 2027 "bounded" and "retirement-only." The Learn page says it includes security and CVE response, documented best practices, and long-term support enablement on a case-by-case basis. The Community Hub post adds help with support incidents that involve supported HPC Pack components and documented functionality.

The exclusion list is longer and more telling. In Argon Systems' copy of the announcement, excluded items include validation or certification of new operating systems, hardware platforms, drivers, firmware, third-party software, or dependencies and migration planning, migration execution, or migration consulting services. The Learn page separately rules out:

• Bug fixes that aren't security-related
• Performance, reliability and scalability improvements
• New features
• Customer-specific hotfixes or code changes
• Design or behavior changes
• Troubleshooting beyond what the existing documentation covers

The two documents word these limits a little differently. Taken together, the practical meaning is clear: if a new Windows Server build, NIC driver or third-party solver breaks your cluster, Microsoft isn't committing to certify or fix it. Security fixes are still promised, and Microsoft decides which ones are needed.

Section summary: Until the deadline you can expect security fixes and help based on existing documentation. Performance work, compatibility testing and migration help are excluded.

The HPC Pack 2016 complication​

The retirement also affects anyone still on an older release. Microsoft's HPC Pack 2016 migration guide says HPC Pack 2016 will be retired on January 12, 2027. That guide recommends one path: the alternative to using HPC Pack 2016 requires using HPC Pack 2019.

That upgrade isn't simple. According to the same guide, there is no in-place upgrade between two major versions of HPC Pack. To install the new version, you need to remove the old one. This will erase your previous jobs and nodes data.

Following that advice now means rebuilding onto 2019 in early 2027 and then leaving 2019 about seven months later. For most HPC Pack 2016 shops, the sensible move is to skip the 2019 step and go straight to a long-term platform. That's our analysis rather than Microsoft guidance, but the dates support it.

For reference, Microsoft's HPC Pack GitHub repository lists the most recent release as HPC Pack 2019 Update 3 Refreshed (6.3.8328) - 2/11/2025. Clusters on older 2019 builds should consider moving to that release while security support still applies.

Who is affected​

Microsoft's Learn guidance says you are affected if you:

1. Run on-premises or hybrid clusters on HPC Pack 2019
2. Use HPC Pack's job scheduler, cluster manager or deployment tooling
3. Rely on HPC Pack for production or research workloads

The Community Hub announcement talks about HPC Pack in general. The detailed Learn migration page covers HPC Pack 2019 specifically. Engineering simulation farms, financial risk grids and university research clusters built around Windows have the most to consider.

Where to migrate​

The Community Hub post names Azure Batch as Microsoft's supported platform for managed HPC and parallel workloads, and adds that customers can choose other Azure services where they fit better. The Learn migration guide lists three options:

• Azure CycleCloud: deploys and manages HPC clusters in Azure with familiar schedulers such as Slurm, PBS and Grid Engine. It automates provisioning and scaling and supports hybrid setups. Microsoft marks it as the recommended choice for HPC orchestration.
• Azure Batch: a fully managed service for large-scale parallel and batch jobs, with autoscaling compute pools and container support. You don't manage scheduler infrastructure.
• Azure Virtual Machines: rebuild a custom cluster on HPC-optimized HB, HC or ND series VMs, running your own or open-source schedulers alongside high-performance storage and networking.

None of these is a drop-in replacement. HPC Pack users typically rely on Windows-native job submission, SOA-style workloads and existing admin scripts. Most of CycleCloud's scheduler options come from the Linux world. Batch changes the model from running a cluster to submitting jobs to a service. Rebuilding on VMs keeps the most control, but you run everything yourself.

Microsoft's Learn guidance says to review your scheduler configuration, workload types, storage dependencies and network requirements before choosing. Its FAQ also warns that migration timelines vary depending on cluster size, workload complexity, and dependencies.

There's one more consideration. All three official options run in Azure. Organizations that must stay fully on-premises for data sovereignty, air-gapped networks or cost reasons won't find an on-prem Microsoft successor in this guidance. They'll need to look at third-party or open-source schedulers themselves. Microsoft's guidance doesn't cover that case.

Section summary: Batch is the managed option, CycleCloud suits teams that want a traditional scheduler, and VMs offer full control. All three run in Azure, and none replaces HPC Pack one-for-one.

A practical migration checklist​

Based on Microsoft's recommended steps, here is a workable order:

1. Inventory. List every HPC Pack cluster, its version and build, and its head, broker and compute nodes, including any hybrid Azure nodes.
2. Map workloads. Record which applications run, how jobs are submitted (GUI, PowerShell, API or SOA clients), and which license servers, file shares and databases they depend on.
3. Sort out HPC Pack 2016 first. If you still run it, the January 12, 2027 deadline comes before the main one. Export your cluster configuration and job history before making any changes.
4. Pick a target. Choose between Batch, CycleCloud and VMs based on how much scheduler behavior you need to keep.
5. Pilot. Move one non-critical workload end to end and compare runtimes, costs and failure handling.
6. Engage Microsoft early. The announcement advises contacting your account team or Cloud Solution Architect early. If you need more time, the Learn page links an extension-of-support questionnaire. It's a request, not a guarantee that an extension will be granted.
7. Finish before August 27, 2027. Treat the earlier of the two published dates as the real deadline.

Not the same as the 2024 Burst to Cloud retirement​

This announcement isn't a repeat of an earlier one. In 2024, Microsoft retired only HPC Pack's cloud-bursting feature. Microsoft's end-of-support list for 2024 includes "HPC Pack burst to Cloud Services" alongside Azure Cloud Services (classic) and other classic services, with an August 31, 2024 retirement date. The lifecycle page gives the same deadline as September 1, 2024 in Pacific Time. That feature depended on the classic Cloud Services platform, which was shut down.

This time the whole product is being retired: the scheduler, the cluster manager and the deployment tooling.

The bottom line​

HPC Pack has had a long run as Microsoft's way to run HPC clusters on Windows Server. Now it has a firm end date, a year of security-focused support, and a clear signal from Microsoft to move the work to Azure. Your clusters will keep running after August 2027, but without further updates or technical support from Microsoft.

For admins, the order of work is straightforward: get off HPC Pack 2016 before January, inventory everything else, and start pilot migrations early enough that last-minute problems don't threaten the deadline. Whether Microsoft's Azure options work for organizations that can't move to the cloud is the part it hasn't addressed.