En
Français
Deutsch
Immersion offers you a new article. Our expert opinion on Nvidia's CloudXR, an innovative technology that aims at improving the performance of virtual reality applications by using the computing power of dedicated graphic stations.
As a user, opt for a fluid and immersive experience. That's one of the benefits we've demonstrated through our testing of the Meta Quest 2 headset.
We tested it for you. We'll show you the results.
If you want to go further, or if you have a use case, don't hesitate to contact us for its implementation!
Streaming for extended reality: towards new horizons
Cloud and virtual reality (VR) are two technological fields that have been developing for several years. Meta, HTC, Pico... all are looking for the standalone headset that will revolutionise the deployment of augmented and virtual reality. These have become powerful tools for the design, production and maintenance of products.
Today, it is necessary to use a headset to access these virtual worlds. However, stand-alone VR headsets such as Meta le Quest 2 still have significant technical limitations. Indeed, the processes involved require significant computing power. The generations of virtual environments require a lot of resources from graphics cards and processors. It is normal to find a powerful graphics card in a PC, but it is impossible today to embed the same type in a headset.
Nvidia, a leader in computing and artificial intelligence, has tried to put the cloud at the service of virtual reality by creating "CloudXR".
The promise? To allow VR headsets to benefit from the computing power of its graphics cards. The headsets would then become simple content displays.
The main objective of CloudXR is to accompany an enrichment of the applications currently embedded in the autonomous headsets.
Figure 1: How NVIDIA CloudXR works (Source : https://cutt.ly/WC1fkNa)
One of Immersion's experts has tested for you the CloudXR from NVIDIA. Testimony and performance evaluation of this technology.
Objectives of the tests? Put NVIDIA CloudXR to the test with one of the most used XR headsets: the Meta Quest 2. Detailed presentation and performance analysis of the Cloud XR e.
The first part will aim to present the experiment in order to facilitate understanding. In a second part, the article will focus on the technical aspects and the graphical performance.
CloudXR: A simple experiment and concrete questions.
In order to carry out his tests, our expert used the following equipment:
Hardware :
- A virtual reality headset: Meta Quest 2 (Find the product sheet on our Cornershop: https://bit.ly/3hovz89 )
- A wifi 6 router to take advantage of a wifi optimised for objects and autonomous headsets
- a PC/server: Nvidia GeForce RTX 3060, Intel i7-9750H, 16GB RAM
Software:
- A Unity application developed by Immersion R&D (described below)
- SteamVR: A tool developed by Valve and HTC to enjoy VR on the hardware of your choice
- OVR Metrics Tools: a headset performance measurement tool developed by Meta.
To create a cloud, you need a server and a network. Servers are powerful computers, especially in the case of virtual creations. In order to be exploited to the full, they must be linked to suitable network infrastructures. The wifi 6 standard is optimised for objects and autonomous headsets.
The OVRMetricsTools application allows the observation and collection of more than 60 headset metrics. Our expert has selected 23 of these metrics, which can be grouped into 3 subgroups: GPU data (Graphics Processing Unit), CPU data (Central Processing Unit) and user experience data (frames per second, battery, headset temperature, etc.). A persistent display (overlay) allows the evolution of the metrics to be seen live, and they are also saved in .csv files for further processing.
Figure 2: Headset field of view: OVRMetricsTools persistent overlay
Testing CloudXR: development and limitation.
In order to test the limitations of NVIDIA CloudXR, our expert developed an application in Unity. This application exists in two forms. The first version was deployed directly on the headset, as is standard procedure when developing with the Meta Quest 2. The second is a Windows application. It is launched on the remote server. The Cloud XR service is the link between the headset and the server. If there are two versions of the application, both run the same scenario.
The scenario is designed to cover a wide range of possible complexities for virtual scenes. A virtual scene is composed of polygons. Each one can have a particular texture and even move in space. The hypothesis is that to make these parameters evolve to increase the complexity of the scene and thus its graphic generation. This requires high computing power in line with the promise offered by CloudXR.
The scenario is divided into three categories. Each category allows the simulation of a particular scene complexity. The polygons used here are cubes.
o First category: The cubes have a simple texture (1st image in figure 3). This configuration simulates the easiest scenes to generate.
o Second category (2nd image in figure 3): The texture of the cubes evolves. Here, they are "glass". The play of light and shadow is then important. The simulated scenes are therefore more complex than the previous category.
o Third category (3rd image in figure 3): Each face of each cube is a video. Each video is independent of the others. This specificity is the most complex. It consumes the most resources.
Figure 3: 100 white cubes, 100 glass textured cubes, 10 video cubes
To refine the measurements in each category, iterations generate 1 cube, then 10 cubes, 100, 1,000, then 10,000 cubes. Each iteration accurately tests the limits of the Meta Quest 2 headset (alone) and the headset with CloudXR.
The tests were conducted under similar conditions. Each test lasted approximately 9 minutes and included the sequence of the 3 categories.
In order to obtain trends for the metrics, 20 tests for each version of the application were performed. Averaging the data of each metric over 20 samples allows to smooth out the faulty measurements and to obtain more reliable data.
Our expert developed a program to automate all data processing. This program extracts all measurements, windows, averages and extracts major trends for each metric. This tool facilitated the analyses that follow later in the article.
What advantages does NVIDIA's CloudXR offer us?
After an initial analysis of the measurements carried out, our expert notes that CloudXR allows us to envisage major advances in mixed reality applications.
Indeed, the use of CloudXR is beneficial for the headset, we quickly observe :
- A decrease in the battery consumption of the headset of about 33%.
- A drop in temperature in the same proportion.
- No additional latency.
- Perceived stability of performance.
When the headset needs to display 10,000 video cubes (a very complex scene), the local version of the application suddenly stops. The Cloud version is more suitable for this generation. Not all videos are displayed correctly, but the application can still be used smoothly. For large amounts of polygons to handle, the headset loses on average 75% more FPS (frames per second) than the local version. The losses are visually disturbing, with the headset dropping from 72 FPS to around 20.
For complex scenes, the advantage is clearly with the CloudXR. On the other hand, for simple scenes or scenes with a few textured polygons the use of the Cloud is not necessarily justified. Indeed, when scenes have up to 1,000 cubes, the local application succeeds without any problems in managing the generations.
Our expert recommends the use of NVIDIA's CloudXR when using complex applications or requiring increased headset portability. The development of 5G and Wi-Fi 6 will allow CloudXR to be used as a fast, reliable and efficient way to use virtual and mixed reality.
"NVIDIA's CloudXR is delivering on its promise and this is just the beginning," concludes our expert.