
For U.S. enterprises, the choice shapes application performance, storage behavior, scalability, operational overhead, and cloud spend. Waste is a real pattern: Flexera's 2026 report estimates 29% of cloud spend is wasted, though that figure covers IaaS and PaaS broadly, not Azure VMs alone.
This guide defines VM series, walks through the main families, compares their trade-offs, and gives you a practical selection process.
Key Takeaways
- Series are workload-oriented families, so choose by resource behavior, not vCPU count.
- B-series suits bursty workloads; D-series covers balanced production needs.
- E/M, F, L, N, and H target memory, compute, local storage, GPU, and HPC demands.
- Baseline CPU, memory, disk, network, region, licensing, and pricing before committing.
What Are Azure Virtual Machine Series?
A series (or family) is a group of VM sizes with similar hardware and features, built for a particular type of workload. Each size inside it fixes a combination of vCPUs, memory, storage capability, and network limits.
Series, size, and SKU
The terms overlap, so keep them straight:
- Family or type: the intended workload category, such as memory optimized.
- Series: sizes sharing similar hardware, such as Dsv5.
- Size: a specific vCPU and memory configuration.
- SKU: the full offering record, which determines generation, disk support, and price.
Names encode features. Lowercase letters flag details: a means AMD, p means Arm, d means local temporary disk, s means Premium SSD compatibility.
Where VMs fit
Azure VMs host application servers, databases, web tiers, dev environments, analytics, AI, and specialized computing. Managed disks and networking are separate design decisions, and they have their own limits.
Why Are Azure VM Series Important?
The series sets your ceilings for latency, throughput, database responsiveness, and scale. Disks and VMs each carry IOPS and throughput caps, and Microsoft's performance documentation says exceeding either one causes capping.
A concrete example: the D8ds_v5 allows up to 16 remote disks, 12,800 uncached IOPS, and 290 MBps on Premium SSD. A fast disk attached to that VM still can't beat those numbers.
Oversized versus undersized
- Oversized: idle capacity and unnecessary spend.
- Undersized: CPU, memory, disk, or network bottlenecks.
The first pick isn't permanent
Demand shifts. Azure Advisor resize recommendations look at CPU, memory, and outbound network over a 7-day lookback, so treat sizing as a recurring task: monitor, test, rightsize.

Azure VM Series Explained
Azure offers families for different workload patterns rather than one universal VM. Names, generations, regions, and retirement dates change, so confirm current details in Microsoft documentation before deploying.
A-series and B-series: entry-level and burstable
A-series is mostly a lifecycle story now. Basic and Standard Av1 retired August 31, 2024, and Av2/Amv2 retires November 15, 2028, so plan migrations.
B-series banks credits while running below baseline, spends them above it, and throttles to baseline when they run out. Good fits:
- Dev and test environments
- Small databases and repositories
- Proof-of-concept builds
- Workloads with long idle periods
Skip it for sustained high CPU or latency-sensitive production. Spot VMs also exclude B-series.
D-series: general purpose
D-series balances CPU, memory, storage, and networking for common production work. Variants differ: Dv5/Dsv5 have no local temp disk, Ddsv5 includes local storage, and Dasv6 has none. Dv3/Dsv3 retire November 15, 2029.
It suits application servers, web workloads, enterprise software, and small to mid-sized databases when telemetry shows balanced demand. Always check the VM-level storage and network limits. Attached disk tier alone does not set overall performance.
E-series and M-series: memory optimized
E-series offers higher memory per vCPU for relational databases, caches, and analytics. Easv6 spans 2 to 96 vCPUs and 16 to 672 GiB. M-series goes much larger: M12s_v3 offers 12 vCPUs and 240 GiB, and Mdsv3 sizes reach 30,400 GiB.
Trade-offs: higher cost, licensing implications, and the risk of buying memory you don't use. For SAP HANA, review each SKU's certification rather than assuming a family is covered.
F-series: compute optimized
F-series favors CPU over memory. Fsv2 runs 2 to 72 vCPUs with 4 to 144 GiB; Fasv6 runs 2 to 64 vCPUs with 8 to 256 GiB and no SMT.
Use it for batch processing, builds, high-throughput app components, rendering, and simulation. Limited memory per vCPU makes it a poor match for databases and caches.
L-series: storage optimized
L-series targets high local I/O: NoSQL, data warehousing, and large transactional systems. The catch is durability. Lsv3 and Lasv3 local NVMe is ephemeral. A successful in-place reboot preserves data; deallocation, redeployment, host failure, and some maintenance moves don't.
Replicate or back up data off the local device, and never treat it like a persistent managed disk.
N-series: GPU-enabled
- NC: compute, AI/ML. NCads H100 v5 offers up to 2 NVIDIA H100 NVL GPUs.
- ND: large deep learning and generative AI. ND H100 v5 has 8 H100 GPUs.
- NV: graphics, CAD, visualization. NVadsA10_v5 uses NVIDIA A10, from fractional to multi-GPU.
Expect higher cost, driver requirements that vary by series, and quota limits. Quota does not guarantee capacity, and GPU stock varies by region, so test deployment early.
H-series and other specialized families
HB, HX, and HC target HPC: CFD, finite element analysis, EDA, molecular dynamics, and weather modeling. HBv4 and HX use 400 Gb/s NDR InfiniBand. HBv2 and HC retire May 31, 2027, and NP-series FPGA retires the same day.

Confidential computing families (DC and EC) use AMD SEV-SNP or Intel TDX. Check support for your exact configuration.
How to Choose the Right Azure VM Series
The right series follows workload behavior, performance goals, resilience needs, and total cost. Not the newest or biggest VM.
- Baseline the workload. Collect CPU, memory pressure, disk IOPS and throughput, network, latency, peak demand, and concurrency. Host metrics need no setup; guest memory needs an agent.
- Classify the dominant constraint. Balanced, burstable, CPU-bound, memory-bound, I/O-bound, GPU, or HPC. That narrows the family.
- Check the full SKU. Verify vCPU-to-memory ratio, processor generation, premium disk support, local storage, max data disks, disk throughput, network bandwidth, and OS compatibility.
- Check region and capacity. Run
az vm list-skusfor location, zone, and subscription restrictions. Regional and family vCPU quotas must both allow the deployment. - Model cost. Use the Azure Pricing Calculator with OS, license, instance count, and runtime.
- Validate and review. Load test, then use Azure Monitor, Azure Advisor, tagging, and budgets to catch idle or mis-sized VMs.

Pricing levers
- Reservations: 1- or 3-year commitments for matching compute; they exclude Windows software, networking, and storage.
- Savings plans: 1- or 3-year hourly dollar commitment, flexible across eligible resources.
- Spot: no SLA, eviction with 30 seconds' notice.
- Hybrid Benefit: Windows Server and SQL Server licenses with active Software Assurance remove the OS charge.
Factor in block storage
VM sizing is only part of the bill. Where block storage drives cost, Lucidity addresses that layer without replacing VM sizing.
AutoScaler expands, shrinks, and optimizes disks with zero downtime across AWS, Azure, and Google Cloud. Lumen analyzes IOPS, throughput, latency, and cost to recommend disk-tier changes, such as moving from Premium SSD to Standard SSD.
What to Check Before Finalising an Azure VM Series
What to Check Before Finalizing an Azure VM Series
- Skip premium or specialized families when telemetry shows general-purpose or burstable is enough.
- Weigh sustained CPU use, memory pressure, local-storage durability, disk throughput, GPU availability, licensing, quotas, and regional capacity.
- Don't size from familiarity or one utilization snapshot. Confirm long-term cost, performance, and resilience, and plan continuous rightsizing.
Conclusion
Azure VM series give you workload-oriented choices, from burstable and general-purpose to memory, compute, storage, GPU, and HPC options. Better outcomes come from matching the full configuration to measured requirements, validating it in your target U.S. region, and watching cost and performance after launch.
Frequently Asked Questions
What are the types of VMs in Azure?
Main categories are general purpose, burstable, compute optimized, memory optimized, storage optimized, GPU accelerated, and high-performance computing. FPGA-accelerated and confidential families also exist.
What are the Azure D series virtual machines?
D-series are general-purpose VMs with balanced compute and memory. They suit application servers, web workloads, and databases. Verify the exact SKU's disk and network limits.
What is the difference between Azure B-series and D-series VMs?
B-series fits low-baseline workloads with occasional CPU bursts and throttles when credits run out. D-series delivers more consistent performance for many production workloads.
Which Azure VM series is best for memory-intensive databases?
E-series covers many databases; M-series handles exceptionally large memory needs. Decide using memory footprint, engine requirements, storage throughput, licensing, and measured behavior.
How do I choose the right Azure VM size and series?
Baseline CPU, memory, disk, network, latency, and peak demand. Then validate the candidate with load testing, Azure Monitor, and cost analysis.
Do Azure VM series affect Azure pricing?
Yes. Family, size, processor generation, OS, region, attached storage, network usage, and commitments all affect cost. Check current Azure pricing rather than relying on generic comparisons.