Modernizing Microsoft SQL Server Performance on Azure with Edsv6 and Intel® Xeon® Processors

Authors:

Janardhana Yoga Narasimhaswamy, Cloud Software Development Engineer, Intel

Swathi Kovvuri, Cloud Software Development Engineer, Intel

For years, running SQL Server on Azure Virtual Machines meant navigating unavoidable trade-offs, oversizing compute to satisfy I/O demands, or compromising between cost efficiency and performance. As workloads scaled and concurrency increased, storage and infrastructure limits often dictated architectural decisions rather than application needs.

With Azure Edsv6 virtual machine families, powered by 5th Generation Intel® Xeon® processors, those compromises are rapidly disappearing. By delivering balanced improvements across compute, memory, and I/O, Edsv6 unlocks a new performance envelope for SQL Server in the cloud.

In this blog, we explore how Azure Edsv6 enables fast, more scalable, and more predictable SQL Server performance, and how it influences the management of mission-critical database workloads on Azure, powered by Intel® Xeon® Processors.

Unlocking SQL Server Performance and Cost Efficiency

As SQL Server workloads grow in size and concurrency, the bottleneck is no longer just compute; instead, it depends on how well CPU, memory bandwidth, memory locality, and storage I/O work together. This makes infrastructure choices a key factor in both performance and cost efficiency. That’s where Azure’s Edsv6 series virtual machines, powered by 5th Generation Intel® Xeon® Scalable processors, represent a significant advancement for running Microsoft SQL Server in the cloud.

Compared to earlier-generation Intel-based Azure virtual machines, Azure Edsv6 provides a more balanced system architecture, enabling SQL Server to effectively convert improvements in compute, memory bandwidth, and I/O into higher throughput and more consistent enterprise-scale performance.

Performance Highlights

To measure the impact of this infrastructure modernization, we assessed SQL Server performance across both OLTP and DSS workloads by comparing two Azure VM generations:

• Edsv6 series, powered by 5th Gen Intel® Xeon® Scalable processors
• Edsv5 series, based on 3rd Gen Intel® Xeon® Scalable processors

 Figure 1 : Performance comparison between E64_dsv5 vs E64_dsv6

Benchmark results show Azure Edsv6 consistently outperforms Edsv5 at the same vCore count across both transactional and analytical workloads. Compared to Edsv5, Edsv6 offers approximately 1.28 times higher OLTP transactions per second and about 1.27 times higher DSS query throughput, demonstrating more efficient per-vCore performance and more predictable scaling for SQL Server workloads.

The conclusion is straightforward: Azure Edsv6, powered by 5th Gen Intel® Xeon® Scalable processors, allows SQL Server to perform more useful work per vCore, enhancing scalability, responsiveness, and overall cost efficiency, especially for mission-critical database workloads that need both performance and predictability, making it an appealing upgrade option for customers modernizing their SQL Server environments on Azure.

Accelerating SQL Server with Storage Innovations in Azure Edsv6 Series

The shift from Azure Edsv5 to Edsv6 series virtual machines marks a fundamental change in how SQL Server manages I/O-intensive workloads in the cloud. With improved Azure Boost, which offloads storage and networking tasks to dedicated hardware, I/O throughput can be significantly increased without using CPU resources.

Edsv6 VMs achieve up to 400,000 IOPS and 12 GB/s of remote disk throughput, marking a significant improvement over Edsv5. Coupled with NVMe-based interfaces for both local and remote disks, Edsv6 consistently offers low latency and high-bandwidth storage that meets the needs of modern, high-concurrency SQL Server deployments. Since storage throughput and latency directly impact query execution, log processing, and tempdb behavior in SQL Server, this architectural update removes a long-standing bottleneck.

SQL Server in Action: Transitioning from I/O Bound to Compute-Driven Performance

In real-world scenarios, SQL Server I/O-intensive benchmarking under sustained, high-concurrency I/O pressure nearing the VM’s storage capacity limits shows that Edsv6 platform improvements with storage provide real performance gains for SQL Server. Importantly, this comparison is made at the same 64 vCore configuration, isolating platform efficiency from scale-up effects.

Compared to the similar Edsv5 64-vCore setup, the Edsv6 64-vCore platform offers:

• 1.60× higher OLTP transaction rate
• 2.41× higher DSS query throughput (queries per hour)

 Figure 2: IO Intensive workload performance comparison between E64_dsv5 vs E64_dsv6

These gains emphasize a fundamental shift: workloads that were previously I/O-bound on Edsv5 are increasingly becoming compute-driven on Edsv6, allowing for higher sustained concurrency and more efficient CPU usage.

This shift is clearly shown in Windows Perfmon and SQL Server waitstats telemetry. These signals confirm that Edsv6 moves SQL Server execution from waiting on infrastructure to performing useful work, allowing predictable and efficient scaling at the same vCore count.

Windows perfmon telemetry

The Standard_E64ds_v6 VM significantly surpasses the Standard_E64ds_v5 VM by providing approximately 63% higher IOPS for OLTP workloads, around 2.3 times greater I/O bandwidth for OLAP workloads, and enhanced overall CPU utilization efficiency.

Table 1: Windows CPU and Disk perfmon telemetry

SQL Server waitstats telemetry

• Much lower read I/O stalls: PAGEIOLATCH_SH decreases by approximately 66%, indicating fast and more consistent data access, which reduces read latency and improves memory and prefetch behavior.
• Significantly decreased in-memory contention: LATCH_SH drops by around 68%, showing enhanced concurrency on shared SQL Server structures as I/O pressure and hot‑page contention are lessened.
• Healthier parallel execution: CXCONSUMER decreases by about 55%, meaning parallel workers synchronize more effectively with reduced coordination overhead.

Collectively, these results show that Azure Edsv6 fundamentally changes SQL Server’s wait profile by removing major infrastructure bottlenecks, especially storage throughput latency and latch contention. As a result, SQL Server on Edsv6 approaches CPU limits more frequently than I/O limits. Workloads that previously depended on storage throughput on Edsv5 are now more compute-driven on Edsv6, resulting in better CPU utilization, higher sustained concurrency, and more consistent performance at scale.

Conclusion

Azure Edsv6 series virtual machines represent a significant advancement for deploying SQL Server at scale in the cloud. By integrating 5th Generation Intel® Xeon® Scalable processors with Azure Boost accelerated I/O, Edsv6 delivers performance, scalability, and predictability once reserved for high-end on-premises systems.

These improvements are not solely due to hardware advances. They result from a long-standing collaboration between Intel and Microsoft to optimize SQL Server performance across on-premises and cloud environments. Instead of a generic hardware–software integration, engineers from both companies co-optimize the SQL Server engine with each major SQL Server release and new Intel® Xeon® architecture.

The impact of this collaboration is evident throughout the entire stack:

• SQL Server engine enhancements optimized for modern Intel® Xeon® processors.
• Scheduler, NUMA, and memory management improvements aligned with the latest Intel® Xeon® platform topology.

In practice, Azure Edsv6 enables SQL Server to take full advantage of Intel® Xeon® architectural enhancements, delivering scalable and predictable performance that allows customers to run mission-critical workloads in Azure with qualities similar to on-premises environments.

For more references

• https://techcommunity.microsoft.com/blog/azurecompute/announcing-general-availability-of-azure-dlde-v6-vms-powered-by-intel-emr-proces/4376186
• https://learn.microsoft.com/en-us/azure/virtual-machines/sizes/memory-optimized/edsv6-series?tabs=sizebasic
• https://www.intel.com/content/www/us/en/partner/showcase/microsoft/overview.html
• https://azure.microsoft.com/en-us/partners/directory/intel-corporation

Testing Configuration

Online Transaction Processing (OLTP) Workload

To evaluate transactional performance, an OLTP workload derived from the TPC Benchmark™ E (TPC-E) was used. TPC-E is widely recognized as a modern OLTP benchmark that represents the behavior of current, enterprise-grade transaction systems. The benchmark simulates financial transaction flows commonly found in industries that depend on high-volume, low-latency processing, such as banking, trading platforms, and order-driven business systems. TPC-E is measured in transactions per second (tpsE). The MSTPCE benchmark tool (v. 1.14.0) was used to build and generate an OLTP workload. The database sizes used for testing are 800,000 and 1,200,000 customers. Although the workload is based on the concepts and structure defined in the TPC E specification, it does not adhere to the full set of TPC compliance and auditing rules. Consequently, the resulting performance data must not be compared with any officially published TPC-E results.

Decision Support and Analytics Workload

Analytical performance was assessed using workload derived from TPC Benchmark™ H (TPC H), a decision-support benchmark that measures how well a system handles complex business queries over large datasets. Designed to challenge systems with complex, ad hoc SQL queries that scan and join large volumes of data to answer business-critical questions. The benchmark workload consists of 22 parameterized queries that represent a supply chain-style data model and reports results using the Composite Query-per-Hour (QphH@Size) metric, which reflects overall query-processing efficiency by accounting for database size. The decision support workload was generated using the MSTPCH benchmark tool (v. 3.0.1) with scale factors of 1 TB and 3 TB. DSS workloads are derived from, but not compliant with, the official TPC-H benchmark specification. Therefore, the observed results are not suitable for direct comparison with any published TPC-H benchmark disclosures.

More information on the benchmark: https://www.tpc.org/default5.asp

Notices and Disclaimer

Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available ​updates. See backup for configuration details. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.