[List of Test Results Collected with EUC Score Experiments]
Benchmarking different aspects of Citrix Virtual Desktop environments and the HDX protocol. The test results presented here are a subset of the complete data sets collected during Citrix test sequences.
If you want to learn how the EUC Score results were produced then check out EUC Score Test Methodology and EUC Score Toolset.
Compare Microsoft RDP versus Citrix HDX on a remote desktop connection to a minimal Windows 365 Enterprise VM with 2 vCPUs and 8GB of memory, hosted on Azure West Europe. This "minimal" or entry-level Windows 365 VM size is very popular among customers or individual users who are trying out Windows 365 Cloud PC for the first time.
Simload and Link | View and Type | Observation | |
---|---|---|---|
![]() |
SL1-AcrobatReaderScroll | SxS View - STD | Almost similar user experience and performance metrics. That's not a big surprise as the GDI calls created by Adobe Acrobat Reader don't require many resources and the user-initiated screen refreshs are triggered once per second only. |
![]() |
SL1-BSPBlendingDX11 | SxS View - STD | This DirectX 11 application is very demanding. As a result, the CPU goes to 100% right from the beginning. The user tries to maneuver in the 3D scene every second, but the system cannot keep up, neither on the RDP nor on the HDX side. |
![]() |
SL1-ChromeCarVisualizer | SxS View - STD | The Car Visualizer WebGL animation is very demanding. The system performance is represented by the rotation speed of the car - which is similarly slow on both sides, despite the good image quality. No remoting protocol can compensate the lack of CPU resources. |
![]() |
SL1-ChromeFishbowlHTML5 | SxS View - STD | On this VM type, the Fishbowl HTML5 app running in the Chrome browser brings the CPU to 100%. The user presses the + button to create a new fish every half second. This works well on the right HDX side (color-coded in red), but it doesn't work on the left RDP side (yellow). |
![]() |
SL1-ChromePhotoGalleryJS | SxS View - STD | The Photo Gallery Simload is based on the resource-intensive Ken-Burns algorithm, showing image zoom, rotation and transition effects implemented in JavaScript. The HDX side shows better usage of the CPU resources and smoother photo transitions compared to the RDP side, with only small differences regarding network consumption. |
![]() |
SL1-ChromeWaterWebGL | SxS View - STD | The Water WebGL Simload running in the Chrome browser brings the CPU of this VM type constantly to 100%. There is no visible difference between the RDP and the HDX sessions. Slightly higher network consumption by HDX, most probably due to the higher frame rate. |
![]() |
SL1-GoogleEarthDX | SxS View - STD | The Google Earth DirectX 9 Simload brings the CPU of this VM type constantly to 100%. For both RDP and HDX, this leads to a very low refresh rate of the animation and massively delayed response times in the Open dialog box. |
![]() |
SL1-MSEdgeAquariumWebGL | SxS View - STD | The WebGL animation of an aquarium with 500 fish shows an unusal behavior on the left RDP side, with a view from the outside of the aquarium sphere. The right HDX side shows the correct animation, but with a lower frame rate. |
![]() |
SL1-MSEdgeFishbowlHTML5 | SxS View - STD | The Fishbowl HTML5 app running in the Microsoft Edge browser brings the CPU of this VM type constantly to 100%. The RDP session on the left produces a much higher CPU queue length and struggles with the user interaction. It fails to create new fish while this is working much better in the right HDX side. |
![]() |
SL1-MSEdgeGifScroll | SxS View - STD | When opening an HTML page with animated GIFs in the Microsoft Edge browser, the left RDP session opens some GIFs with a delay. Scrolling the page up and down works equally well on both sides. |
![]() |
SL1-MSEdgeVideoConf6 | SxS View - STD | Running six MP4 videos simultaneously in the Microsoft Edge browser show a similar good quality for both RDP and HDX, bringing the CPU to 100%. There is a notable difference in the user session-specific CPU load, but this doesn't influence the system-wide CPU consumption. |
![]() |
SL1-MSEdgeVideoGrid | SxS View - STD | When running nine MP4 videos simultaneously in the Microsoft Edge browser, the HDX session show a better user experience. There are more studders and a higher network consumption on the RDP side. |
![]() |
SL1-MSOPptSimple | SxS View - STD | This entry-level Windows 365 VM type is well suited for Office workloads such as the PowerPoint Simple Simload, independently of the remoting protocol. The only (technical) benefit that the HDX session brings to the table is the higher frame rate. |
![]() |
SL1-MSOWordGif | SxS View - STD | This entry-level Windows 365 VM type is well suited for Office workloads such as the Simload showing a Word document with GIF images, independently of the remoting protocol. The only (technical) benefit that the HDX session brings to the table is the higher frame rate. |
![]() |
SL1-RollercoasterDX9 | SxS View - STD | Only eight or nine frames per second of the Rollercoaster DirectX 9 animation are rendered in the frame buffer due to CPU limitations. The output on the client side is studdering and shows less than 1fps, independently of the remoting protocol. It's notable that RDP requires significantly less bandwidth and CPU after 25 seconds. |
![]() |
SL1-WMPlayer1080pMP4 | SxS View - STD | Running a 1080p MP4 video in the Windows Media Player fails completely on the left RDP side and show unacceptable smearing effects on the right HDX side. |
![]() |
SL1-WMPlayer720pMP4 | SxS View - STD | When running a 720p MP4 video in the Windows Media Player, everything looks equally good on both sides as long as the video player is in windowed mode. But when switching to full-screen mode, the video freezes on the RDP side. And there is a much lower CPU demand on the HDX side. |
![]() |
SL3-AppDialog | SxS View - STD | App Dialog score: 0.36 on the RDP side, 0.32 on the HDX side (lower is better). Only small differences when measuring the time is takes to open a dialog window. |
![]() |
SL3-AppStart | SxS View - STD | The RDP session shows an almost 20% better score when starting an application. App Start score: 0.78 on the RDP side, 0.93 on the HDX side (lower is better). |
![]() |
SL3-FractalsDragon | SxS View - STD | RDP beats HDX when comparing their GDI+ Dragon scores. GDI+ Fractals Dragon Curve score: 2.64 on the RDP side, 2.91 on the HDX side (lower is better). |
![]() |
SL3-FractalsPythagoras | SxS View - STD | RDP beats HDX when comparing their GDI+ Pythagoras scores. GDI+ Fractals Pythagoras Tree score: 4.4 on the RDP side, 4.89 on the HDX side (lower is better). |
![]() |
SL3-GDIPlusRect | SxS View - STD | RDP beats HDX when comparing their GDI+ Rectangles scores. GDI+ Rectangles score: 2.04 on the RDP side, 2.71 on the HDX side (lower is better). But the screen output is studdering significantly more on the RDP side. |
![]() |
SL3-IOPS | SxS View - STD | Surprisingly, RDP beats HDX by approx. 5% when comparing their IOPS scores - even though it's the identical underlying storage system. IOPS score: 13.83 on the RDP side, 14.64 on the HDX side (lower is better). |
Everbody who expected a big performance boost when using Citrix HDX instead of Microsoft's RDP side-by-side protocol stack on an entry-level Cloud PC will be disappointed. This tests shows test scenarios where HDX performs better and test scenarios where RDP performs better - but many times there is no visible difference. The lack of CPU resources and low storage throughput is more significant on both sides. What's notable is that the frame rate is 30fps on the RDP side while it is 60fps on the HDX side - resulting in higher network consumption for HDX. The entry-level Windows 365 VM type under test is only suited for simple applications in a task worker scenario. Things may look very different when using larger Windows 365 Cloud PC VM types with more CPU, memory, storage and possibly GPU resources.
RDP Connection to Cloud PC
HDX Connection to Cloud PC
System Under Test: Azure West Europe, Windows 365 Enterprise, Windows 11, AMD EPYC 7763 2vCPUs @ 2.45GHz, 8GB RAM, 128GB Microsoft Virtual Disk, no GPU, Remote Desktop Services SxS Network Stack.
System Under Test: Azure West Europe, Windows 365 Enterprise, Windows 11, AMD EPYC 7763 2vCPUs @ 2.45GHz, 8GB RAM, 128GB Microsoft Virtual Disk, no GPU, Citrix HDX WS x64 7.33.3000.29.
Connection: Microsoft RDP-UDP, 12ms round trip time.
Connection: Citrix HDX, 12ms round trip time.
Endpoint: NUC2, Intel NUC 8i7HNK with Windows 11, i7-8705G CPU @ 3.10GHz, 16GB RAM, 500GB Samsung SSD 850 EVO M.2, AMD Radeon RX Vega M GL, 4GB VRAM, Windows 365 App 1.3.192.0.
Endpoint: NUC2, Intel NUC 8i7HNK with Windows 11, i7-8705G CPU @ 3.10GHz, 16GB RAM, 500GB Samsung SSD 850 EVO M.2, AMD Radeon RX Vega M GL, 4GB VRAM, Citrix Workspace App 2311.
Explore the perceived user experience of a HDX connection from the EUC Score lab in Germany to a GPU-accelerated Citrix session hosted on Azure West Europe.
Simload and Link | View and Type | Observation | |
---|---|---|---|
![]() |
SL1-GoogleEarthDX9 | Single View - GPU | Outstanding performance of Google Earth over HDX, significant GPU usage. |
![]() |
SL1-MSOPptSimple | Single View - GPU | User interacting with simple PowerPoint slide deck. Only minimal GPU usage. |
Great performance of the Azure NCasT4 v3 VM over Citrix HDX.
Compare an RDP connection to the physical reference lab machine Lancelot with a HDX connection to a GPU-accelerated Citrix session hosted on Azure West Europe.
Simload and Link | View and Type | Observation | |
---|---|---|---|
![]() |
SL1-ChromeFishbowlHTML5 | SxS View - GPU | Performance is almost identical, but HDX connection requires less network bandwidth when compared to RDP-UDP connection. |
![]() |
SL1-MSEdgeVideoConf6 | SxS View - GPU | Loading the videos takes longer in the HDX connection, then the performance almost identical. The HDX connection requires less network bandwidth. |
![]() |
SL1-DominoOpenGL | SxS View - GPU | HDX connection with great OpenGL performance, but it shows some latency effects. In addition, it requires more network bandwidth when compared to the RDP-UDP connection. |
![]() |
SL1-FurMarkOpenGL | SxS View - GPU | OpenGL application uses 100% of the GPU capacity, resulting in 75 fps in the RDP-UDP connection and more than 100 fps in the HDX session. RDP requires more network bandwidth in the beginning before the protocol adapts. |
![]() |
SL1-MSEdgeCarVisualizer | SxS View - GPU | Car rotates significantly faster in the HDX session while required network bandwidth stays significantly lower. |
![]() |
SL1-WMPlayer720pMP4 | SxS View - GPU | Video performance is identical on both sides, but HDX connection requires lower network bandwidth when compared to RDP-UDP connection. |
The RDP connection to the physical lab machine Lancelot with NVIDIA M5000 is outperformed by the Azure NCasT4 v3 VM over Citrix HDX in many use cases, except when IOPS are relevant.
RDP Connection to Lancelot
HDX Connection to Azure NCasT4 v3 VM
System Under Test: Physical Lab Machine Lancelot, Windows 10, i7-11700K CPU 3.60GHz, 8 physical cores (= 16 threads), 64GB RAM, Crucial MX500 SSD 1TB, NVIDIA Quadro M5000 GPU, 8GB VRAM.
System Under Test: Azure West Europe, NCasT4 v3 VM type, Windows 10 Pro, AMD EPYC 7V12 64-Core Processor 2.445Ghz 8vCPUs, 28GB RAM, HDD 126.45GB, NVIDIA Tesla T4, 16GB VRAM.
Connection: RDP-UDP, 0ms round trip time.
Connection: Citrix HDX, 20ms round trip time.
Endpoint: NUC2, Intel NUC 8i7HNK with Windows 11, i7-8705G CPU @ 3.10GHz, 16GB RAM, 500GB Samsung SSD 850 EVO M.2, AMD Radeon RX Vega M GL, 4GB VRAM.
Endpoint: NUC2, Intel NUC 8i7HNK with Windows 11, i7-8705G CPU @ 3.10GHz, 16GB RAM, 500GB Samsung SSD 850 EVO M.2, AMD Radeon RX Vega M GL, 4GB VRAM.