Table of Contents

• Difference in Azure VM SKUs
• Introduction OBUX
• Setup and configuration
• Hypothesis and results
• OBUX System Score
• OBUX User Score
• Cost evaluation
• Performance data
• Conclusion

Microsoft Azure offers a wide selection of virtual machine SKUs, each representing a distinct configuration tied to underlying hardware. While many users default to Intel-based instances, AMD-based SKUs have emerged as a competitive alternative. This GO-EUC research investigates the performance and cost implications of selecting AMD-based virtual machines in Azure compared to their Intel-based counterparts.

Difference in Azure VM SKUs

Before examining the results, it is essential to understand the different virtual machine categories available in Microsoft Azure. Azure offers a wide range of SKU types, each tailored to specific workload characteristics. These include:

• General purpose;
• Compute optimized;
• Memory optimized;
• Storage optimized;
• GPU accelerated;
• FPGA accelerated.

Based on the naming convention, these SKUs can be differentiated as it is built up in the following way.

1. Typically, families are represented by a single letter, although some, such as GPU sizes (ND-series, NV-series, etc.), use two letters.
2. Subfamilies are usually denoted by a single uppercase letter, but some, like the Ebsv5-series, are still classified under their parent family due to differences in features.
3. If a CPU doesn’t list a feature letter, it defaults to Intel x86-64 CPUs. An AMD CPU will be indicated with an “a,” while an ARM-based CPU (like Microsoft Cobalt or Ampere Altra) will use “p.”
4. Size names can contain numerous extra features, ranging from none (as in the Dv5-series) to multiple (such as the Dplds_v6-series).
5. Spacers can appear several times in a size name, as seen in the ND_H100_v5-series, where they separate the GPU ID from the rest of the name.
6. Version numbers are included in size names when there are multiple versions of the same series. Often, the first version of a series (such as the HB-series or B-series) does not include the version number.

For more information, refer to the Microsoft documentation.

Azure VM SKU series often includes multiple CPU configurations under a single naming convention. While Microsoft does not disclose usage statistics, the D-series is widely regarded as one of the most commonly deployed SKUs. As such, this research focuses specifically on the D-series to evaluate differences between Intel and AMD-based instances.

The following table contains a breakdown of the differences between the Intel and AMD-based SKUs in the same series:

* Microsoft does not explicitly state which AMD EPYC 9004 series processor they use, but based on the collected data, it appears to be an AMD EPYC 9V74.

As learned from a previous research, upgrading to a newer version of the same SKU is beneficial. To explicitly see the differences between the SKU versions, it is recommended to read the research here.

Introduction OBUX

The Open Benchmark User Experience (OBUX) mission is to establish an open, vendor-neutral standard for benchmarking user experience—free to use, simple to execute, and designed to scale alongside evolving technology needs. Built and maintained by a passionate community of EUC professionals, OBUX is grounded in real-world testing and engineered with the future in mind.

The benchmark scoring is fully open source, released under the permissive MIT license. This openness invites participation and scrutiny, ensuring continuous evolution and trust. It’s built with flexibility in mind—capable of running across a wide range of environments, from physical desktops and laptops to virtual machines in VDI, DaaS, and Cloud PC infrastructures.

While OBUX is not directly associated with GO-EUC, in this and future researches, OBUX will be used to measure the user experience alongside the other tool set. Learn more about OBUX here.

Setup and configuration

The primary objective of this research is to evaluate the performance differences between Intel- and AMD-based SKUs in Microsoft Azure. The D2 series is selected, as it is the default when creating a new virtual machine in Azure. This VM contains 2 vCPUs with 8GB of memory. The following SKUs are included:

• Standard D2s v5
• Standard D2as v5
• Standard D2s v6
• Standard D2as v6

Cost information is collected from the Microsoft website using the following specifications:

All virtual machines are provisioned via Terraform in the Azure East US 2 region, due to the availability of the required SKUs, and are configured using Ansible. The virtual machines run standalone and are accessible using Remote Desktop Protocol (RDP) via a public IP address protected by a network security group. This means that there is no brokering involved, such as with Azure Virtual Desktop, for serving the machines.

The virtual machines are running Windows 11 24H2 Enterprise and are optimized using the Citrix Optimizer. OBUX version 85 was used for this research, which was initiated from the command line using the “run once” registry with a 2-minute idle period preceding it. The logon process is not part of the OBUX benchmark, so custom automation is used to set up a Remote Desktop session to the machine running in Azure.

Each OBUX test cycle is repeated 10 times to ensure consistency, and averages are used in this research, like any other research at GO-EUC. Additional performance data is collected using Telegraf and stored in an on-premises Influx database.

Hypothesis and results

In the comparison between Intel and AMD, the hypothesis is that Intel outperforms AMD, which is based on personal experience. Now that being said, this is not done with complete confidence, as there is a clear shift in the market towards AMD, as Intel is losing market share.

Based on previous research, it is expected to see an increase in performance between the different SKU versions.

OBUX System Score

The OBUX system score is an evaluation of the machine’s available performance. The score is based on specific CPU-intensive calculations, disk IO operations, and application start times. The measurements are based on timing, so the longer it takes, the lower the overall score will be. The higher the score, the better.

In the most recent generation SKU, the v6, the performance difference between AMD and Intel offerings is negligible. The Standard_D2s v6 shows an OBUX score of 709, while the AMD counterpart, the Standard D2as v6, has a score of 708.

For the previous generation, the performance gap is more evident. The AMD-based SKU, the Standard D2as v5, shows a score of 610, while the Intel SKU, Standard D2s v5 manages to show a score of 561, which is around 9% lower.

Both the Intel-based SKUs and the AMD VMs show a substantial increase in performance over the generations, with Intel showing a 26.4% increase in performance from v5 to v6 and AMD showing a 16.1%.

OBUX User Score

The OBUX user score is a scoring mechanism used to evaluate the user experience. This measurement is taken on the virtual machine side, so it does not account for the remoting protocol. The score is based on the response times in the application, as measured by the APDEX methodology. Scoring is evaluated from 0, which represents an unacceptable user experience, to 1, which represents an excellent user experience, with higher being better.

The UX results show a slightly different result than with the System Scores.

All SKUs show a UX score above 0.89, very close to the perfect UX score of 1.0. This suggests that, from an end user’s perspective, applications should feel very responsive and provide an excellent user experience.

In a direct comparison between each generation, the Intel-based SKUs outperform the AMD SKUs, with a 1.7% higher UX score in the v6 generation and 1.3% higher scores for the v5 generation.

Cost evaluation

When running in a public cloud, cost is an important factor. Optimizing the cost against usability could be beneficial for your organization. The following comparison is the prices per SKU per month.

It shows a price difference between Intel and AMD, favoring AMD. Based on the OBUX system score, in combination with price, it makes sense to prefer an AMD-based SKU, as this is a better balance in price and performance.

The difference is 11% in favor of AMD. In context, when you host an Azure Virtual Desktop environment based on 1000 machines, this is a direct cost saving of $7.373,- per month, which is $88.476,- yearly. This can make a significant difference in your organization, so this is something to consider, as the virtual machine SKU can be easily switched. Note: The virtual machine must be stopped before the SKU can be changed.

Performance data

The following data is collected using Telegraf, which was running during the OBUX benchmark. It is important to understand that the performance data is also an average of the 10 individual runs.

On the CPU level, the AMD-based SKU has a slightly lower footprint compared to Intel. This difference will not have a direct impact, so it would not have a significant effect on a single-user scenario.

During the OBUX system score runs, the storage performance metrics were also collected as part of the data collection.

On the reads/sec there is a notable difference at the start of the test, where the v6 series consistently has a higher reads/sec. While this was consistent for each run, it cannot be directly clarified what the root cause is, and that requires further investigation. On the writes/sec the differences are minimal.

The memory footprint is overall the same. Based on the raw dataset, there is evidence that leads to the suspicion that the v6 series is quicker in clearing the memory, resulting in a higher footprint for the v5 series.

Conclusion

Azure offers a broad range of virtual machine SKUs, from general-purpose to GPU-accelerated, supporting diverse workloads. Among these, the D-series, commonly based on Intel processors, is frequently used in VDI and DaaS deployments.

Based on OBUX benchmark results, it can be concluded that AMD-based SKUs offer a better balance of performance and cost. The v6 SKU version has a minimal difference.

In terms of cost, the difference is 11% in favor of AMD. When running Intel-based SKUs, consider using AMD and realize a significant cost difference while maintaining the same overall performance and user experience.

Now, there may be a specific requirement for Intel-based virtual machines, but if you have the opportunity, it’s recommended to switch. But, also keep in mind that some specific SKU quotas might be limited in your region; it is therefore recommended to check the quotas on your Azure subscription.

Do you prefer Intel or AMD-based machines in your environment? Let us know in the comments below.

Photo by Jatin Gajjar on Unsplash