【Gamystery】EP20 | Testing Google Immersive Stream for XR: From Installation Hurdles to Over $2,400 in Costs—Is It Worth It?
- 彥澤 廖
- Jan 22
- 4 min read
Updated: Feb 3
Before the Chinese New Year New Year, I had a free week and decided to adjust my writing approach. EP19 and EP20 are both products of this new approach. These two articles target different audiences as a trial run. The “Introductory Series” takes more time to prepare (worked on intermittently over the week) because explaining abstract concepts in detail relies heavily on past real-world experience. Merely researching or conducting short-term investigations won’t easily produce engaging content.
In contrast, writing content for special topics is faster: just a bit of screenshot-taking and note-recording during work accelerates the writing process. The material I’m sharing today took about one or two workdays of testing. The main criterion for choosing topics in this series is “what is needed for work and development.” Whenever a project in hand has research value, I’ll use my spare time or partial work hours to explore it.
From my past to present attempts at new technologies, most have turned out somewhat disappointing or have notable flaws in actual usage. I hope this special-topic series helps you save time when selecting the right technology to use.
This post shares the process and technical challenges of using one of Google GCP’s services, Immersive Stream for XR.
Background
I happened to come across a Google-official cloud streaming service. Around the same time, my project was using Unreal Engine's official Pixel Streaming solution, yet I felt it wasn't the optimal architecture for our project.
From the official explanation, the use cases look quite good, offering cross-domain integration and solving multi-platform issues.My initial impression was that it seemed extremely well-done. Well... that's not surprising, considering it's the official video.
Installation and testing process
Important links:
The official instructions are quite detailed; simply follow the recommended steps. You can run this sample without installing or launching Unreal Engine.
You only need to install Unreal Engine if you want to delve deeper into AR or add additional project resources. Besides opening the Unreal Engine editor, the main goal is to compile and test a Linux version locally so you won’t end up waiting a long time only to have the build fail after uploading.
Additional Note: On GitHub, I've seen some applications—like automotive or interior design demos—where you need to open the project, go into Project Settings, and change the map (Level) settings for the correct content to appear.
Based on my experience, I want to supplement details that the official documentation doesn't fully explain, including how to troubleshoot certain obstacles.
Region settings are extremely important—choosing the wrong region can lead to deployment failures.
I recommend starting with the GUI approach as outlined in the official documentation, and only using the command line for uploading content. If you see the message “Script execution is disabled on this system” after entering the official command, please refer to this tutorial to resolve it.
Those are all the issues I encountered during my testing.
Finally, after spending an entire day at home, I ended up with a logo that's bigger than my desk.
Postscript
Overall, the deployment isn' t particularly difficult. Apart from the somewhat lengthy deployment time, there aren't many challenging steps; once you've done it once, it becomes quite straightforward. I also opened the Unreal Engine project to look at the Blueprint settings (as shown in the official instructions). It seems not too different from a typical game or interactive application. In more advanced scenarios, you could collect and use information from connected users to track and improve the user experience.
The main downside of Immersive Stream for XR is that it's extremely expensive. I only realized this when I checked the GCP billing console while writing this article. It charges daily, and one machine costs about 100 USD per day; about a month has passed from my initial testing to drafting this article. In December alone, it consumed $2,400 of my free credits. As of January, the total has accumulated to over $50,000—truly shocking.
Initially, I thought you'd only be billed when a user connects to use the service, but it turns out that as soon as the service is deployed and active, the billing starts. (I'm not sure if this is considered obvious, but the pricing documentation is quite vague.) In practical terms, it's probably only viable to rent it for very short-term events. If you're looking to operate a long-term service, it's definitely not recommended (no wonder so few people are using it).
If I manage to adjust the billing limits, I'll update you on what happens next (currently, the Gemini bot isn't working—so sad...). The worst-case scenario is that my credit card fails to pay, causing the project to be suspended.
If you enjoyed this content, please show your support by giving us a like on Facebook.