, yesterday Nvidia announced new Titan RTX. How can you compare it with Titan V and which one will be better for GPU Lightmass? Titan V is stronger in FP32, but Titan RTX in FP64 and other specs. What is more important for GPU lightmass?
And second question, Titan RTX has NvidiaLink support, do you have plans to implement it? Previously you said that SLI will not be supported, but maybe NvidiaLink is more important?
thank you so much for your hard work. Finally with the turing support the lightmass is baking at 1Grays/s. Just one PC is now faster then previously 6 computers using swarm with CPU lightmass. This will save tens of hours. Awesome job!
While GPULightmass supports Turing cards now, it is not using hardware ray tracing (RTX). Instead it uses the traditional CUDA cores. The reason why I choose to do so is to support RTX the architecture of GPULightmass must be fully rewritten and will not be compatible with old cards unless I keep two ray tracing engine in the code base (the DXR fallback layer is horribly slow and MS have abandoned it)
Needed your opinion on something. So as you might know recently Nvidia launched the RTX Titan 24 GB graphics card. Now given that money’s not an issue, should I buy the RTX 6000 which costs twice as more than the RTX Titan or should I go with the RTX Titan because their specs are literally the same !!
I work in real estate and do Architectural Visualization demos. So GPU rendering is of huge importance to me.
Please let me know what’s the difference between these two cards and which one is best for GPU rendering based Arch Viz demos.
Also, would I benefit from two RTX Titans using SLI and can Unreal use memory from both cards or it’s just core sharing?
Your elaborate comments will really be appreciated !!
Hi, please don’t be too lazy and try to read what I asked coupe posts ago! And here is his answer:
AFAIK, SLI is not and will be not supported. NVLink is still an open quesiton. GPU Lightmass using for now only CUDA cores, so new Titan RTX meanwhile will not be using for 100%, but 24 GB VRAM is very important for us, so probably it will be better investment for the future.
I just installed and tried GPU Lightmass for the first time with 4.21.
Speed and quality are great, but my result is much darker than with the CPU baker. It is as if less GI bounces are computed. I tried increasing the indirect lighting contribution of the lights without any difference in the result. Bounces in the lightmass settings are set to 100.
Also I get a lot of errors in the lightmaps when setting any rectangle light to stationary. Only when setting it to static the bake works fine.
Time to get a GPU with more VRAM ! I also got this error on big projects (multiple houses and buildings). Happens when swarm unloads everything in the memory of your GPU : if you dont have enough, the baker crashes with this error.
Try to increase the FireflyClampingThreshold in the BaseLightmass.ini in the engine\config folder from 10 to 50 or 100. Be aware that since you increase the threshold you might need to increase samples as well to get a clean result.
Thanks! I´ll check it out! I already increased the samples in the BaseLightmass.ini and the results were great. I got a better quality in 10 minutes on my 1080ti than in almost two hours on my CPU.
But I didn´t touch the fireflyClampingThreshold until now, because I didn´t know what it was for.
Thats funny, for me GPU luightmass is like 10 x brighter than the simple CPU lightmass. I was using 300 lumen lights and i had to put them down to 30 lumen to match my old results.
(the DXR fallback layer is horribly slow and MS have abandoned it)
I also got this error on big projects (multiple houses and buildings). Happens when swarm unloads everything in the memory of your GPU : if you dont have enough, the baker crashes with this error.