When I render my image, it stops, giving the error message "CUDA ERROR: Launch failed in cuCtxSynchronize() Line 1372", then when I try to start the render again, I get the error message "cuCtxCreate: Launch failed" all the answers I have seen say to update your graphics card driver, however, mine is up to date. I have a GTX1080 with 4gb of RAM.
4
-
$\begingroup$ Most likely you ran out of memory, what's the peak number you see in the info above the image? $\endgroup$– Johan ApesMar 7, 2018 at 15:47
-
$\begingroup$ My peak memory while rendering is 963M but on the Blender taskbar it shows as 2265M $\endgroup$– ConlanMar 7, 2018 at 22:29
-
$\begingroup$ Do you run other programs that require VRAM simultaneously? This ways my first idea, but there are lot of variables that can be causing this in your scene, complexity of the geometry, size of the textures. A quick solution besides simplyfing your scene is to use smaller tile size, like 64,32 or even 16. Also restarting the program helps to free some memory. blender.stackexchange.com/questions/96878/… blender.stackexchange.com/questions/94178/… $\endgroup$– Johan ApesMar 8, 2018 at 9:36
-
$\begingroup$ try updating ur driver $\endgroup$– Dragion TechNov 7, 2018 at 22:30
Add a comment
|
2 Answers
$\begingroup$
$\endgroup$
Bumped into the same error once.
Found following thread at developers forum:
Then it occurred to me that the Gforce Experience runs in the background (and being in my opinion, not a well written program) may be overlapping the memory of Blender or at least hogging the memory. So I un-installed it and magic! All Cuda errors and memory errors went away. Blender works again as it should!
$\begingroup$
$\endgroup$
I used to have this error, till i lower my render tile sizes to a value, that is lower than my output image size, my output image was "1920*1080" so i made my tile size 500 by 500.
and if you have the option "square samples" ticked, then untick it.
From my experience the problem is from either the tile size or the square samples or both
hope that works.
