Principal Cloud Solutions Architect
Azure compute updates give teams more VM choices and new ways to standardize configurations, handle capacity constraints and recover from zone failures. They matter to Azure administrators, infrastructure teams and application owners who provision VMs or run workloads that can use Arm or interruptible capacity. John Savill's [MVP] discusses these Azure changes in an Ignite-focused update.
The updates span VM hardware, reusable configuration and operational resilience. Azure Cobalt 200 Arm VMs are generally available, joining VM families based on other processor platforms. The new Cobalt 200 families include Dp, Ep, Mp and Lp, and Savill describes them as offering up to 50% CPU improvement over Cobalt 100 and 20% better remote-storage IOPS.
For teams creating VMs repeatedly, Compute templates for Azure offer a way to save versioned configurations in Azure Compute Gallery. A template can include operating-system information, storage, networking and security settings. It can then be referenced when creating a VM, with specific values overridden when a deployment needs a different setting.
Several previews address deployment and recovery challenges. Adaptive placement can use an equivalent VM when the requested size is unavailable, while zone resiliency lets an operator manually recover a zonal VM to an alternate zone after a zone failure. Compute Spot Plus targets workloads that need more runtime consistency than regular Spot, with a stated target of at least 60 minutes.
Cobalt 200 is relevant to organizations evaluating Arm-based VMs for workloads that fit its VM families. The update identifies Dp, Ep, Mp and Lp families, but choosing a VM still means matching the workload to the appropriate family and validating that its software can run on the Arm architecture. Teams already using Infrastructure as Code with Bicep or Terraform may have less need for Compute templates, since they may already maintain reusable VM definitions that way.
Compute templates are in limited preview. Adaptive placement, zone resiliency and Compute Spot Plus are also described as previews, so organizations should treat them as evaluation options rather than assume they are ready for every production deployment. The Azure Compute Fleet integration with IBM Spectrum LSF is generally available; it combines Fleet's VM provisioning with LSF's high-performance computing job scheduling.
For administrators, the main decision is whether a capability fits an existing deployment pattern. Templates are aimed at gallery-based VM configuration reuse; adaptive placement and zone recovery address specific infrastructure scenarios; Spot Plus is for workloads that can tolerate eviction. The Azure compute topic covers related infrastructure updates.
Compute templates provide a repeatable path from a saved VM configuration to a new deployment. Rather than rebuilding the same settings each time, an operator references a versioned template and changes individual values only when necessary. The template is stored in Azure Compute Gallery.
Adaptive placement addresses a different problem: a specific VM size may not be available in the chosen location. The feature can deploy an equivalent VM based on factors such as virtual CPU count and performance, helping avoid deployment failure due to that capacity constraint. Zone resiliency is a manual recovery mechanism: when a zonal VM's zone fails, an operator can initiate recovery to another zone while retaining the disks and configuration.
Compute Spot Plus builds on the principle behind Spot VMs: use spare Azure capacity for workloads that can be evicted when regular workloads need that capacity. The Plus option aims to provide at least 60 minutes of runtime and gives more eviction notice. If eviction occurs before the full 60 minutes, the workload is not charged for that 60-minute period, according to the update.
Preview status matters when deciding whether to depend on these features. Compute templates are specifically identified as a limited preview, and the placement, recovery and Spot Plus capabilities are also previews. Teams should check whether each preview fits their deployment requirements before relying on it for critical workloads.
Templates are not necessarily a replacement for Infrastructure as Code. Savill notes that teams using Bicep or Terraform may already define reusable VM configurations there; templates offer an Azure Compute Gallery-integrated alternative. Spot Plus also does not remove eviction risk: it remains best-effort, costs more than regular Spot, and is still aimed at workloads able to handle interruption.
Adaptive placement may select an equivalent rather than the exact requested VM size, so teams should consider whether that substitution is acceptable for their workloads. Zone recovery is described as a manual action, not automatic failover. Cobalt 200's Arm architecture is another reason to check workload compatibility before adopting a VM family.
Yes. Cobalt 200 Arm VMs are generally available, with Dp, Ep, Mp and Lp families.
They store versioned VM configurations in Azure Compute Gallery. When creating a VM, you reference a template and can override specific values.
The described preview lets an operator manually initiate recovery to an alternate zone after a zone failure, keeping the VM's disks and configuration.
Spot Plus targets at least 60 minutes of runtime and provides more eviction notice, but remains best-effort and costs more than regular Spot.