So for testing purposes, I got my hands on a Titan X today, and found a strange result when it came to performance. In the sci-fi hallway, in both the 4.8 and 4.9 VXGI builds, I noticed that for some reason, in mode (was on Windows 7 so wasn’t an option), in 4.8 VXGI it was running much, much faster.
At 1440p, MapSize 32, 4 cones, sparse tracing 4, specular disabled:
4.8 VXGI-
10.96ms total
1.75ms VXGI WS
2.15ms VXGI SS
4.9 VXGI-
22.4ms total
12.19ms VXGI WS
2.5 VXGI SS
I went back and forth between the two engine builds for over an hour, trying to make absolutely sure the settings were the same, but no matter what I did, I could not get 4.9’s build to run anywhere close to what 4.8 was putting out (And for all intents and purposes, the two builds visually looked almost the same. Just a tiny bit different in the color correction settings or something between them, but nothing that would cause another 10ms of frame time).
Has anyone else noticed a slowdown in the 4.9 build? I know the official changelog says " GPU features are not supported in mode yet" but that shouldn’t be impacting performance so badly.
Thank you for noticing . There is nothing wrong with your setup or configuration, and there is a performance regression. And guessed right that it’s caused by refactoring: the new NVRHI backend works through UE RHI and it turns out that way is much more expensive on the CPU side. I didn’t anticipate that, so didn’t benchmark that change. Sorry about that.
I’ll fix in the next few days.