
The catch: the published P10 spec doesn't equal what your application gets. VM limits, caching, I/O size, concurrency, and burst behavior all change the result.
This guide covers the current P10 size, IOPS, throughput, limits, pricing, measurement, and resizing, based on current Microsoft documentation.
Key Takeaways
- P10 provisions 128 GiB, 500 IOPS, and 100 MB/s, with credit-based bursting available.
- Performance is tied to the disk tier, so more speed means a different tier or disk type.
- Your real ceiling is the lowest limit among disk, VM, caching, and workload.
- Decide using weeks of measured data, not one peak reading.
What Azure P10 Managed Disk Size and IOPS Represent
P10 is a Premium SSD performance tier. Microsoft lists it at 128 GiB, 500 provisioned IOPS, and 100 MB/s. Microsoft's disk types page is the source of record.
Units matter. Microsoft's table uses GiB (1 GiB = 1,073,741,824 bytes), though Azure property names still say diskSizeGB. Also, 128 GiB is the provisioned size, not the free space your OS reports after formatting.
The three metrics
- IOPS: I/O requests completed per second.
- Throughput: data read or written per second.
- Latency: total time one I/O request takes to complete.
These are provisioned service characteristics, not a promise that every workload will hit them.
Factors That Influence P10 Performance in Real-World Operation
Requests that exceed either the disk or the VM allocation get capped. Several factors decide which limit you hit first.
Configuration constraints
- VM uncached limits: cap IOPS and bandwidth across all attached disks.
- Host caching:
None,ReadOnly, orReadWrite. Cache-served reads don't count against disk limits.ReadWriterisks data loss on a VM crash if the application doesn't persist writes correctly. - Disk count: more disks don't lift the VM's aggregate ceiling.
- Redundancy: ZRS writes have higher latency than LRS.
Workload characteristics
- Read/write mix: ratio and random versus sequential access shape effective throughput
- I/O size: operations above 256 KiB count as multiple I/Os
- Queue depth: more pending requests can raise IOPS, but too many worsen latency
- Burst and concurrency: spikes and parallel clients hit caps sooner than steady load
Filesystem overhead, backups, background jobs, and seasonality also push field results away from a short synthetic test or a datasheet.
Range and Key Technical Properties of Azure P10
P10's behavior is set by its SKU, configuration boundaries, and workload. Check current values on Microsoft Learn rather than older announcements.
Nominal Operating Range
| Tier | Size (GiB) | Provisioned IOPS | Provisioned MB/s | Burst IOPS | Burst MB/s |
|---|---|---|---|---|---|
| P6 | 64 | 240 | 50 | 3,500 | 170 |
| P10 | 128 | 500 | 100 | 3,500 | 170 |
| P15 | 256 | 1,100 | 125 | 3,500 | 170 |
You're billed for the provisioned tier, not occupied filesystem bytes. P6 suits small, light disks. P15 makes sense when sustained IOPS regularly approach 500 or you need more capacity. All three share the same burst ceiling, so the real difference is sustained performance.
Boundary Limits and Bursting
The disk is rarely the only ceiling. Compare two VM sizes:
| VM | Max data disks | Uncached IOPS | Uncached MB/s |
|---|---|---|---|
| Standard_D2s_v5 | 4 | 3,750 | 85 |
| Standard_D4s_v5 | 8 | 6,400 | 145 |
A D2s_v5 caps uncached throughput at 85 MB/s, below P10's 100 MB/s rating. The VM becomes the bottleneck before the disk does.
Bursting: P10 qualifies for credit-based bursting, up to 3,500 IOPS and 170 MB/s for 30 minutes. Credits build when you run below baseline and drain when you burst. It's best-effort and carries no extra charge. On-demand bursting requires disks over 512 GiB, so P10 can't use it.

To spot exhausted burst, watch Data Disk Used Burst IO Credits Percentage and Data Disk Used Burst BPS Credits Percentage. They report credits used every 5 minutes.
Safe Operating Margin and Technical Trade-offs
Leave headroom. If sustained utilization sits near 100% of the disk or VM limit, or queue depth and latency climb together, you're too close to the edge.
IOPS, throughput, and average I/O size are coupled. Throughput ≈ IOPS × I/O size. Large sequential reads can hit 100 MB/s with only a few hundred IOPS. Small random writes can hit 500 IOPS while moving very little data.
That coupling is fixed within each Premium SSD tier. Premium SSD v2 works differently. Capacity, IOPS, and throughput are configured independently, with a baseline of 3,000 IOPS and 125 MB/s. The trade-offs:
- No host caching
- No OS disk use
- LRS only
- Limited regions
You can convert directly from Premium SSD, but it requires a restart and has prerequisites, including caching set to None and no configured bursting. Verify eligibility before you migrate.

How P10 Is Specified, Measured, and Validated
Treat P10's spec as a hypothesis. Live VM and application demand will confirm or refute it.
Specification and Documentation
Microsoft Learn covers disk types, VM limits, bursting, and regional support. Use the Azure pricing page for current costs. To identify a disk's configuration:
- Portal: disk Size + performance blade; VM Disks settings for caching and attachment.
- CLI:
az disk show -g <resource-group> -n <disk>shows SKU (Premium_LRSorPremium_ZRS), size, tier, andmanagedBy. - PowerShell:
Get-AzDisk -ResourceGroupName <rg> -DiskName <disk>. - Caching mode: check
storageProfile.dataDisks[].cachingthroughaz vm showorGet-AzVM.
Provisioning and Resizing Validation
Expansion is two steps: increase the Azure disk size, then extend the partition and filesystem in the guest OS.
- OS disks need VM deallocation. Eligible Premium SSD data disks can expand online.
- Going from 4 TiB or less to above 4 TiB requires deallocation and detachment.
- Guest visibility of the new size can take up to 10 minutes.
- Windows: rescan, then extend via Disk Management or DiskPart.
- Linux: rescan, then grow the partition and filesystem or LVM layout.
Shrinking isn't supported directly. Microsoft's workaround is to create a smaller disk, copy data, verify, and retire the original.
You can also raise P10 performance without adding capacity. A 128 GiB P10 can select P15 through P50 tiers, billed at the selected rate until changed back. Downgrades are limited to once every 12 hours.
Measurement and Verification Methods
Use Azure Monitor to separate disk, VM, and application signals:
- Disk limits:
Data Disk IOPS Consumed Percentage,Data Disk Bandwidth Consumed Percentage - VM limits:
VM Uncached IOPS Consumed Percentage,VM Uncached Bandwidth Consumed Percentage, plus the cached equivalents of each - Pending requests:
Data Disk Queue Depth
Most metrics emit every minute. Microsoft benchmarks with DiskSpd on Windows and fio on Linux. Test representative block sizes, queue depths, read/write mixes, and durations. Warm the cache before cached-read tests.
Validate against both average and peak periods. For FinOps decisions, a representative window of weeks or months beats a single instantaneous reading.

Implications of Operating Outside the Recommended Range
When demand exceeds the disk or VM boundary, requests get capped. Latency rises, application throughput falls, and database operations stretch out. Azure has no separate availability SLA for Disk Storage, and the VM SLA doesn't guarantee P10 burst performance.
The cost of over-provisioning
P10 costs $19.71 per month for LRS and $29.57 for ZRS in East US (Azure managed disk pricing). You pay regardless of occupied space. Idle IOPS and unused capacity are avoidable spend. Compare measured use against P6, P15, Standard SSD, or Premium SSD v2 before you commit.
Waste is common at scale. Flexera's 2026 report estimates 29% of IaaS and PaaS spend is wasted, though that isn't specific to block storage. Lucidity's own research puts average enterprise disk utilization at about 30%, with 8-10% of disks leading to downtime.
Common misinterpretations
- Treating nominal specs as guaranteed application performance
- Assuming a larger disk always removes the bottleneck (the VM may be the limit)
- Confusing 30-minute burst with sustained performance
- Resizing based on one short-lived spike
Those patterns make manual rightsizing hard to sustain across many volumes. Lucidity is an Azure-compatible platform that gives real-time visibility into every volume. Its Lumen tool uses historical IOPS, throughput, latency, and cost trends, rather than point-in-time snapshots, to recommend tiering. AutoScaler can expand and shrink block storage with zero downtime.
Conclusion
P10's published 128 GiB, 500 IOPS, and 100 MB/s govern provisioning. Actual performance depends on the VM, workload, caching, burst rules, and other Azure limits.
Before you finalize a P10 setup:
- Verify current Microsoft specifications
- Measure representative demand
- Keep operating headroom
- Choose a configuration that meets both reliability and FinOps goals
Frequently Asked Questions
What is an Azure P10 managed disk?
P10 is a Premium SSD managed disk tier provisioned at 128 GiB with 500 IOPS and 100 MB/s. Confirm current values in Microsoft Learn.
How many IOPS does an Azure P10 disk provide?
500 provisioned IOPS, with credit-based bursting to 3,500 for up to 30 minutes. VM limits, caching mode, and workload can reduce effective performance.
What are the limits of an Azure P10 managed disk?
The disk caps at 500 IOPS and 100 MB/s sustained, and 3,500 IOPS and 170 MB/s burst. VM uncached limits, disk-count limits, and ZRS region availability also apply.
Can I increase P10 disk performance without increasing its size?
Yes. Select a higher performance tier such as P15 or P20 without expanding capacity, or move to Premium SSD v2 where supported. If the VM is the bottleneck, resize the VM instead.
How much does an Azure managed disk cost?
Cost depends on SKU, size, redundancy, and region. P10 in East US is $19.71 per month (LRS) or $29.57 (ZRS). Use the Azure pricing calculator for current figures.
What sizes are available for Azure managed disks, and how do I resize them?
Sizes depend on disk family. Expand in Azure, then extend the partition and filesystem in the guest OS. Shrinking requires creating a smaller disk and migrating data.