Benchmarks: Citrix vs AVD Performance Comparison¶

Audience: EUC Architects, VDI Engineers, Performance Engineers Purpose: Quantitative comparison of user density, login times, protocol latency, GPU performance, Teams optimization, and printing across Citrix CVAD and Azure Virtual Desktop. Last updated: 2026-04-30

Methodology¶

These benchmarks represent aggregated data from customer migrations, Microsoft internal testing, and published vendor benchmarks. All tests compare equivalent Azure VM sizes running on the same Azure infrastructure. Individual results vary based on workload profile, user behavior, network conditions, and configuration.

Test environment baseline:

• VM size: Standard_D8s_v5 (8 vCPUs, 32 GB RAM)
• OS: Windows 11 Enterprise 24H2 multi-session (AVD) / Windows Server 2022 (Citrix)
• Storage: Azure Files Premium (FSLogix) / Azure Files Standard (Citrix UPM)
• Network: same Azure region, same VNet
• Users: simulated with Login VSI or equivalent load generator

1. User density comparison¶

User density measures how many concurrent users a single VM can support while maintaining acceptable user experience (VSImax or equivalent threshold).

1.1 Knowledge worker profile¶

Workload: Office applications (Word, Excel, Outlook, Edge), Teams, light file operations.

AVD advantage: 40% higher density on the same VM due to lower OS overhead of Windows 11 multi-session vs Windows Server with Desktop Experience.

1.2 Data analyst profile (CSA-in-a-Box)¶

Workload: Power BI Desktop, Azure Data Studio, Python/Jupyter, Edge with 10+ tabs, Excel with large datasets.

1.3 Task worker profile¶

Workload: single application (line-of-business app), minimal multitasking.

2. Login time comparison¶

Login time is the total duration from user credential submission to interactive desktop availability. This is the most visible user experience metric.

2.1 Login time breakdown¶

FSLogix advantage: 10--40 seconds faster login due to VHDx mount (1.5--3s) vs UPM file sync (15--45s).

2.2 Login time by profile size¶

FSLogix login time is nearly independent of profile size because the VHDx mount is a constant-time operation. UPM login time scales linearly with profile size because files are copied at login.

3. Protocol latency comparison¶

3.1 Round-trip time (RTT)¶

Measured between user input and screen update.

HDX retains a 10--15% latency advantage on UDP transports due to deeper protocol optimization. The gap narrows with each AVD update. For most users (LAN and WAN), both protocols provide imperceptible latency.

3.2 Bandwidth consumption¶

Average bandwidth per user during typical knowledge worker activity:

4. GPU performance¶

4.1 GPU VM options¶

4.2 Graphics performance (SPECviewperf 2020)¶

HDX 3D Pro retains a 10--15% advantage for GPU-accelerated graphics due to H.264/HEVC hardware encoding optimizations. For most CAD and GIS workloads, AVD GPU performance is acceptable. For professional visualization requiring maximum frame rates, HDX 3D Pro is still the better protocol.

5. Teams optimization¶

5.1 Teams media offload comparison¶

Both AVD and Citrix support WebRTC media offload for Teams -- video, audio, and screen sharing are processed on the client device, not the session host.

Parity: Teams optimization is functionally equivalent on both platforms. Both use the same WebRTC media engine for client-side processing.

5.2 Zoom optimization¶

6. Printing performance¶

6.1 Print latency¶

Time from print command to first page output:

6.2 Large document printing¶

7. Scalability benchmarks¶

7.1 Host pool scaling¶

7.2 Profile storage IOPS¶

8. Summary¶

Bottom line: AVD provides significantly higher density and faster login times. Citrix retains advantages in protocol optimization for graphics-intensive and extreme-low-bandwidth scenarios. For the majority of enterprise and federal workloads (knowledge workers, data analysts, task workers), AVD delivers better or equivalent performance at substantially lower cost.

9. CSA-in-a-Box data analyst workload benchmarks¶

9.1 Power BI Desktop performance¶

9.2 Azure Data Studio / Jupyter performance¶

9.3 Data analyst desktop density recommendation¶

Based on combined workload testing (Power BI + Azure Data Studio + Edge + Teams):

The E-series (memory-optimized) VMs are well-suited for data analysts who work with large in-memory datasets.

10. Benchmark methodology notes¶

10.1 Test tools used¶

• Login VSI: industry-standard VDI load generation and user experience measurement
• Azure Monitor + AVD Insights: production telemetry for real-world validation
• PerfMon / Performance Monitor: Windows performance counters for CPU, memory, disk, network
• Wireshark + Protocol Analysis: network-level protocol measurement for RTT and bandwidth
• SPECviewperf 2020: standardized graphics performance benchmark

10.2 Caveats¶

• All benchmarks are indicative, not absolute. Individual results vary based on workload mix, user behavior, network conditions, and configuration
• Citrix HDX performance depends heavily on Citrix Workspace app version and policy configuration
• AVD RDP Shortpath performance depends on network path quality and NAT traversal success
• GPU benchmarks depend on driver version and GPU partitioning configuration
• Login times depend on profile size, GPO complexity, and storage IOPS provisioning
• Teams optimization depends on client hardware (CPU, GPU, webcam quality)

10.3 How to run your own benchmarks¶

1. Deploy a pilot AVD environment (see Tutorial: AVD Deployment)
2. Keep your existing Citrix environment running in parallel
3. Use Login VSI or a similar tool to generate comparable load on both platforms
4. Measure: login time, session host CPU/memory, user input delay, protocol RTT
5. Compare results side-by-side
6. Run for at least 5 business days to capture representative usage patterns

Maintainers: CSA-in-a-Box core team Last updated: 2026-04-30