And sadly no sign for any SLI/NVlink support yet as in Octane…
I thought those Titan RTX cards have just more RAM, but not really more GPU horse power when looking at the specs…
And sadly no sign for any SLI/NVlink support yet as in Octane…
I thought those Titan RTX cards have just more RAM, but not really more GPU horse power when looking at the specs…
Did you test it in BF Blender 2.81? It works fine on my side (AI48_002.blend)
Hmm…what is BF blender?
Using 20191121_rtx Linux version on a 1950X Debian 10 system with 64GB RAM…
Ah no…dunno if this Cuda error message is memory related as I have no problems with the AI43 scenes…but they are also much smaller in size texture wise…
But hopefully Mathieu can tell more when he is back from family leave…
BTW: With GPU rendering getting faster and faster the time it takes to precalculate the scene and updating images takes almost as long as the rendnering itself…especially when doing animations…
Would the total time be reduced either with a higher core-count CPU or with a higher clocked CPU?
Edited : I tried again. OptiX rendering works fine (AI48_02) with E-Cycles_2.81_realease_win_rtx_v3 build. I don’t know why not before.
Is rtx_v3 the latest release for Windows? Hard to tell from the the file list on gumroad…
Could try it as well as I have Debian/WIn10 dual boot system…but preferably stick on Debian due to other software/scripts…
Yep, I just confirmed that the AI48_01 scene is also well rendered. and It’s really fast
Weird…no error messages in console?
I always quit blender before trying a new scene just to be sure GPU is newly initialised…
Hmm…in render settings it says 64 samples for final rendering…but shows then 4096 AA samples…why this high number? And what does the appended AA mean behind the samples number?
It works fine on my side. That’s because Square Samples is checked. Uncheck Square Samples and try it on a lower sample.
Just trying AI48_01 on WIndows 10 with latest 2.81_v3_rtx version…
Is dead slow using GPU with 2 * RTX2080Ti and tells me it takes around 5 hours to finish…
Also I see no GPU activity in the task manager:
Your tile size is way too small.
Well I made them smaller just to see a progress earlier (o;
Seems Windows task manager doesn’t show any GPU 3D activity when CUDA is used…
Oddly in those evermotion scenes I can’t set depth-of-field with focus on an object and using F-stop value of 2.0 or lower…everything is focussed…but that is also true for regular blender version…
Well did that…seems the “slow” preset gives the best results as “fast” and “medium” are too bright…well…could adjust exposure…
What puzzles me is that focus-on-obejct has no effect…added a new camera just to be sure…chose the glass jar in the middle as focus object and tried with f-stop 1.4 and 16:
F-Stop 1.4:
F-Stop 16:
Task Manger by default does not show “compute”.
click on the “3d” or other head of the graphs, and change to “compute0” through “compute3” untill one shows actual usage.
option 2 - download GPU-Z, as that will show actual usage.
small tiles have been fixed for Nvidia GPU’s some time ago, so there is “next to no” impact in 32x32 vs 256x256 tile settings.
AMD wise, there is impact.
Found it…
World unit scale was set to 0.01, meaning I had to set it to 1 and scale everything by 0.01…
F-Stop: 1.0
Seems that the DOF is scaled together with the world unit scaling…
After reading this huge thread, I decided to place an order as well…
I was very skeptic, but I got your point and want support you.
Great work!
Just a heads up that @bliblubli is on a well deserved hiatus currently so he may not respond too soon! Let us know if you have any issues though - we can probably help you get them sorted.