Over the weekend I integrated both VXGI and WaveWorks into my UE4 build, made a scene and thought it would be cool to tell and show you guys some result.
The scene contains one directional light, everything is dynamic and rendered in realtime.
It ran at an avg. of ~22 FPS, with min at 18 and max at 24.
Everything was at max, usually referred to as Epic, temporal AA, rendered at my monitor’s native 2560x1440 with a screen percentage at 100%.
I did not use vxgi specular tracing as it turns SSR off and I did want to have that post effect, I did try it but it was a pretty easy visual call at the end especially as it would have cost a few additional frames. Light propagates with multiple bounces.
Recorded with FRAPS ran as a standalone from the Editor (minimized, realtime unchecked) as it was already 2:30am and packaging failed the first try.
This is my machine that this was made and ran at:
Motherboard: Z97-DELUXE(NFC & WLC)
PSU: Cooler Master “V Series” V850
CPU: Intel® Core™ i7-4790K 4.6Ghz
CPU cooler: Hydro Series™ H100i Extreme Performance
GPU: GeForce GTX TITAN X
RAM: G.SKILL Trident X Series 32GB (4 x 8GB) DDR3 2400 Mhz
SSD: Samsung SSD 850 PRO 256GB
HDD: WD Green 2TB
CASE: NZXT. Phantom 410 Black
During development there was on average 9.3GB of vram in usage and about 19GB of ram.
It would be great if everybody would get to see the original video that is 9.42GB in size, compression in this video takes its toll.
I’m planning to release a special Titan X version of a finished demo, not locked to it, just recommended, memory.
I manually toggle between fast/slow day/night cycle during the video.
Comparison, top image is before integration, bottom after.
No problem, since I’m not at home during most of the week, ironically in my day job I work with Unity, I’ve done tests via Teamviewer so take them with a grain of salt, should still hold relatively.
VXGI, with multi-bounce (official description of this is: “Experimental support for multi-bounce indirect illumination, by fetching VXGI irradiance output from the previous frame”): ~21 FPS
VXGI without multi-bounce: ~26.5 FPS
No VXGI: ~78 FPS
WaveWorks performed about 0.5 FPS better than the default ocean material. That could probably be a whole different story if you ran it on AMD GPU, because of the lower tessellation throughput, but my level probably wasn’t big / complex enough to cause any bottlenecks there.
This may sound silly although it was interesting, I found all the code from Nvidia’s builds and ported it over to my code, along with libraries.
Thanks, Haha, no worries man, it’s still a very high end feature it should be some time before we get this in mainstream games.