Hi all, we have been testing CloudXR for a while now and have just realised after testing that CloudXR’s performance largely differs when using a remote desktop protocol such as NICE DCV compared to not using any rdp at all. (CloudXR performs better without the rdp) I am unsure of the reason for this discrepancy in performance and was hoping if anyone knows the reason for this.
For comparison: without rdp, CloudXR averages around 90 fps. with NICE DCV, CloudXR averages around 70-80 fps. Also, what is the limiting hardware in this case, GPU used in this case is a RTX 4080 and is more than capable to encode frames for the stream.
Hi, @stsh_0823!
This kind of cross-interaction is not unexpected, but also not something we’ve tested. There could be a number of reasons for this decrease in performance and some have potential fixes.
- If you have the SteamVR VR view window open on the desktop, this ties SteamVR to the Windows desktop compositor, and with NICE capturing the Windows desktop, there could be a chain of cross-interactions that is slowing down the SteamVR compositor, leading to an actual slowdown of the rendered frames submitted to CloudXR. This is an easy check: close the SteamVR “VR View” window and see if this improves things.
- NICE DCV may be competing with CloudXR over network bandwidth. Both are meant to be real-time low-latency streams, and both do a lot of QoS work. NICE DCV may be prioritizing its traffic higher than CloudXR. The fix here would be to tweak NICE DCV’s settings, or else to switch to a less aggressive remote desktop protocol.
- If the server is on a bandwidth limited final-hop connection, and particularly if it is on WiFi or similar, then NICE DCV might be reducing the available bandwidth for CloudXR. The server should be on a hard-wired gigabit or higher ethernet connection to run both of these services.
Note that these are just ideas. I don’t know exactly what’s going on.